1 /*
2 * producer_avformat.c -- avformat producer
3 * Copyright (C) 2003-2021 Meltytech, LLC
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2.1 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
18 */
19
20 #if !defined(_POSIX_C_SOURCE) || _POSIX_C_SOURCE < 200809L
21 # undef _POSIX_C_SOURCE
22 # define _POSIX_C_SOURCE 200809L
23 #endif
24 #if !defined(_XOPEN_SOURCE) || _XOPEN_SOURCE < 500
25 # undef _XOPEN_SOURCE
26 # define _XOPEN_SOURCE 500
27 #endif
28
29 #include "common.h"
30
31 // MLT Header files
32 #include <framework/mlt_producer.h>
33 #include <framework/mlt_frame.h>
34 #include <framework/mlt_profile.h>
35 #include <framework/mlt_log.h>
36 #include <framework/mlt_deque.h>
37 #include <framework/mlt_factory.h>
38 #include <framework/mlt_cache.h>
39 #include <framework/mlt_slices.h>
40
41 // ffmpeg Header files
42 #include <libavformat/avformat.h>
43 #include <libswscale/swscale.h>
44 #include <libavutil/samplefmt.h>
45 #include <libavutil/pixdesc.h>
46 #include <libavutil/dict.h>
47 #include <libavutil/opt.h>
48 #include <libavutil/channel_layout.h>
49 #include <libavutil/imgutils.h>
50 #include <libavutil/version.h>
51
52 #define USE_HWACCEL 1
53 #if USE_HWACCEL
54 #include <libavutil/hwcontext.h>
55 #endif
56
57 #ifdef AVFILTER
58 #include <libavfilter/avfilter.h>
59 #include <libavfilter/buffersink.h>
60 #include <libavfilter/buffersrc.h>
61 #endif
62
63 // System header files
64 #include <stdlib.h>
65 #include <string.h>
66 #include <pthread.h>
67 #include <limits.h>
68 #include <math.h>
69 #include <wchar.h>
70 #include <stdatomic.h>
71
72 #define POSITION_INITIAL (-2)
73 #define POSITION_INVALID (-1)
74
75 #define MAX_AUDIO_STREAMS (32)
76 #define MAX_AUDIO_FRAME_SIZE (192000) // 1 second of 48khz 32bit audio
77 #define IMAGE_ALIGN (1)
78 #define VFR_THRESHOLD (3) // The minimum number of video frames with differing durations to be considered VFR.
79
80 struct producer_avformat_s
81 {
82 mlt_producer parent;
83 AVFormatContext *dummy_context;
84 AVFormatContext *audio_format;
85 AVFormatContext *video_format;
86 AVCodecContext *audio_codec[ MAX_AUDIO_STREAMS ];
87 AVCodecContext *video_codec;
88 AVFrame *video_frame;
89 AVFrame *audio_frame;
90 AVPacket pkt;
91 mlt_position audio_expected;
92 mlt_position video_expected;
93 int audio_index;
94 int video_index;
95 int64_t first_pts;
96 atomic_int_fast64_t last_position;
97 int video_seekable;
98 int seekable; /// This one is used for both audio and file level seekability.
99 atomic_int_fast64_t current_position;
100 mlt_position nonseek_position;
101 atomic_int top_field_first;
102 uint8_t *audio_buffer[ MAX_AUDIO_STREAMS ];
103 int audio_buffer_size[ MAX_AUDIO_STREAMS ];
104 uint8_t *decode_buffer[ MAX_AUDIO_STREAMS ];
105 int audio_used[ MAX_AUDIO_STREAMS ];
106 int audio_streams;
107 int audio_max_stream;
108 int total_channels;
109 int max_channel;
110 int max_frequency;
111 unsigned int invalid_pts_counter;
112 unsigned int invalid_dts_counter;
113 mlt_cache image_cache;
114 int yuv_colorspace, color_primaries, color_trc;
115 int full_luma;
116 pthread_mutex_t video_mutex;
117 pthread_mutex_t audio_mutex;
118 mlt_deque apackets;
119 mlt_deque vpackets;
120 pthread_mutex_t packets_mutex;
121 pthread_mutex_t open_mutex;
122 int is_mutex_init;
123 AVRational video_time_base;
124 mlt_frame last_good_frame; // for video error concealment
125 int last_good_position; // for video error concealment
126 #ifdef AVFILTER
127 AVFilterGraph *vfilter_graph;
128 AVFilterContext *vfilter_in;
129 AVFilterContext* vfilter_out;
130 #endif
131 int autorotate;
132 int is_audio_synchronizing;
133 int video_send_result;
134 #if USE_HWACCEL
135 struct {
136 int pix_fmt;
137 int device_type;
138 char device[128];
139 AVBufferRef* device_ctx;
140 } hwaccel;
141 #endif
142 };
143 typedef struct producer_avformat_s *producer_avformat;
144
145 // Forward references.
146 static int list_components( char* file );
147 static int producer_open( producer_avformat self, mlt_profile profile, const char *URL, int take_lock, int test_open );
148 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index );
149 static void producer_avformat_close( producer_avformat );
150 static void producer_close( mlt_producer parent );
151 static void producer_set_up_video( producer_avformat self, mlt_frame frame );
152 static void producer_set_up_audio( producer_avformat self, mlt_frame frame );
153 static void apply_properties( void *obj, mlt_properties properties, int flags );
154 static int video_codec_init( producer_avformat self, int index, mlt_properties properties );
155 static void get_audio_streams_info( producer_avformat self );
156 static mlt_audio_format pick_audio_format( int sample_fmt );
157 static int pick_av_pixel_format( int *pix_fmt );
158
159 /** Constructor for libavformat.
160 */
161
producer_avformat_init(mlt_profile profile,const char * service,char * file)162 mlt_producer producer_avformat_init( mlt_profile profile, const char *service, char *file )
163 {
164 if ( list_components( file ) )
165 return NULL;
166
167 mlt_producer producer = NULL;
168
169 // Check that we have a non-NULL argument
170 if ( file )
171 {
172 // Construct the producer
173 producer_avformat self = calloc( 1, sizeof( struct producer_avformat_s ) );
174 producer = calloc( 1, sizeof( struct mlt_producer_s ) );
175
176 // Initialise it
177 if ( mlt_producer_init( producer, self ) == 0 )
178 {
179 self->parent = producer;
180
181 // Get the properties
182 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
183
184 // Set the resource property (required for all producers)
185 mlt_properties_set( properties, "resource", file );
186
187 // Register transport implementation with the producer
188 producer->close = (mlt_destructor) producer_close;
189
190 // Register our get_frame implementation
191 producer->get_frame = producer_get_frame;
192
193 // Force the duration to be computed unless explicitly provided.
194 mlt_properties_set_position( properties, "length", 0 );
195 mlt_properties_set_position( properties, "out", 0 );
196
197 if ( strcmp( service, "avformat-novalidate" ) )
198 {
199 // Open the file
200 if ( producer_open( self, profile, mlt_properties_get( properties, "resource" ), 1, 1 ) != 0 )
201 {
202 // Clean up
203 mlt_producer_close( producer );
204 producer = NULL;
205 producer_avformat_close( self );
206 }
207 else if ( self->seekable )
208 {
209 // Close the file to release resources for large playlists - reopen later as needed
210 if ( self->audio_format )
211 avformat_close_input( &self->audio_format );
212 if ( self->video_format )
213 avformat_close_input( &self->video_format );
214 self->audio_format = NULL;
215 self->video_format = NULL;
216 }
217 }
218 if ( producer )
219 {
220 // Default the user-selectable indices from the auto-detected indices
221 mlt_properties_set_int( properties, "audio_index", self->audio_index );
222 mlt_properties_set_int( properties, "video_index", self->video_index );
223 mlt_service_cache_put( MLT_PRODUCER_SERVICE(producer), "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
224 mlt_properties_set_int( properties, "mute_on_pause", 1 );
225 }
226 }
227 }
228 return producer;
229 }
230
list_components(char * file)231 int list_components( char* file )
232 {
233 int skip = 0;
234
235 // Report information about available demuxers and codecs as YAML Tiny
236 if ( file && strstr( file, "f-list" ) )
237 {
238 fprintf( stderr, "---\nformats:\n" );
239 void *state = NULL;
240 const AVInputFormat *format = NULL;
241 while ((format = av_demuxer_iterate(&state))) {
242 fprintf( stderr, " - %s\n", format->name );
243 }
244 fprintf( stderr, "...\n" );
245 skip = 1;
246 }
247 if ( file && strstr( file, "acodec-list" ) )
248 {
249 fprintf( stderr, "---\naudio_codecs:\n" );
250 void *state = NULL;
251 const AVCodec *codec = NULL;
252 while ((codec = av_codec_iterate(&state))) {
253 if ( codec->decode && codec->type == AVMEDIA_TYPE_AUDIO )
254 fprintf( stderr, " - %s\n", codec->name );
255 }
256 fprintf( stderr, "...\n" );
257 skip = 1;
258 }
259 if ( file && strstr( file, "vcodec-list" ) )
260 {
261 fprintf( stderr, "---\nvideo_codecs:\n" );
262 void *state = NULL;
263 const AVCodec *codec = NULL;
264 while ((codec = av_codec_iterate(&state))) {
265 if ( codec->decode && codec->type == AVMEDIA_TYPE_VIDEO )
266 fprintf( stderr, " - %s\n", codec->name );
267 }
268 fprintf( stderr, "...\n" );
269 skip = 1;
270 }
271
272 return skip;
273 }
274
first_video_index(producer_avformat self)275 static int first_video_index( producer_avformat self )
276 {
277 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
278 int result = -1; // not found
279
280 if ( context ) {
281 unsigned int i;
282 for ( i = 0; i < context->nb_streams; i++ ) {
283 if ( context->streams[i]->codecpar &&
284 context->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_VIDEO )
285 break;
286 }
287 if ( i < context->nb_streams ) {
288 result = i;
289 }
290 }
291 return result;
292 }
293
294 #if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(56, 1, 0)
295
296 #include <libavutil/spherical.h>
297
get_projection(AVStream * st)298 static const char* get_projection(AVStream *st)
299 {
300 const AVSphericalMapping *spherical = (const AVSphericalMapping*) av_stream_get_side_data(st, AV_PKT_DATA_SPHERICAL, NULL);
301
302 if (spherical)
303 return av_spherical_projection_name(spherical->projection);
304 return NULL;
305 }
306
307 #endif
308
309 #include <libavutil/display.h>
310
get_rotation(AVStream * st)311 static double get_rotation(AVStream *st)
312 {
313 AVDictionaryEntry *rotate_tag = av_dict_get( st->metadata, "rotate", NULL, 0 );
314 uint8_t* displaymatrix = av_stream_get_side_data( st, AV_PKT_DATA_DISPLAYMATRIX, NULL);
315 double theta = 0;
316
317 if ( rotate_tag && *rotate_tag->value && strcmp( rotate_tag->value, "0" ) )
318 {
319 char *tail;
320 theta = strtod( rotate_tag->value, &tail );
321 if ( *tail )
322 theta = 0;
323 }
324 if ( displaymatrix && !theta )
325 theta = -av_display_rotation_get( (int32_t*) displaymatrix );
326
327 theta -= 360 * floor( theta/360 + 0.9/360 );
328
329 return theta;
330 }
331
filter_restricted(const char * in)332 static char* filter_restricted( const char *in )
333 {
334 if ( !in ) return NULL;
335 size_t n = strlen( in );
336 char *out = calloc( 1, n + 1 );
337 char *p = out;
338 mbstate_t mbs;
339 memset( &mbs, 0, sizeof(mbs) );
340 while ( *in )
341 {
342 wchar_t w;
343 size_t c = mbrtowc( &w, in, n, &mbs );
344 if ( c <= 0 || c > n ) break;
345 n -= c;
346 in += c;
347 if ( w == 0x9 || w == 0xA || w == 0xD ||
348 ( w >= 0x20 && w <= 0xD7FF ) ||
349 ( w >= 0xE000 && w <= 0xFFFD ) ||
350 ( w >= 0x10000 && w <= 0x10FFFF ) )
351 {
352 mbstate_t ps;
353 memset( &ps, 0, sizeof(ps) );
354 c = wcrtomb( p, w, &ps );
355 if ( c > 0 )
356 p += c;
357 }
358 }
359 return out;
360 }
361
362 /** Find the default streams.
363 */
364
find_default_streams(producer_avformat self)365 static mlt_properties find_default_streams( producer_avformat self )
366 {
367 unsigned int i;
368 char key[200];
369 AVDictionaryEntry *tag = NULL;
370 AVFormatContext *context = self->video_format;
371 mlt_properties meta_media = MLT_PRODUCER_PROPERTIES( self->parent );
372
373 // Default to the first audio and video streams found
374 self->audio_index = -1;
375 int first_video_index = self->video_index = -1;
376
377 mlt_properties_set_int( meta_media, "meta.media.nb_streams", context->nb_streams );
378
379 // Allow for multiple audio and video streams in the file and select first of each (if available)
380 for( i = 0; i < context->nb_streams; i++ )
381 {
382 // Get the codec context
383 AVStream *stream = context->streams[ i ];
384 if ( ! stream ) continue;
385 AVCodecContext *codec_context = stream->codec;
386 if ( ! codec_context ) continue;
387 AVCodecParameters *codec_params = stream->codecpar;
388 AVCodec *codec = avcodec_find_decoder( codec_params->codec_id );
389 if ( ! codec ) continue;
390
391 snprintf( key, sizeof(key), "meta.media.%u.stream.type", i );
392
393 // Determine the type and obtain the first index of each type
394 switch( codec_params->codec_type )
395 {
396 case AVMEDIA_TYPE_VIDEO:
397 // Save the first video stream
398 if ( first_video_index < 0 )
399 first_video_index = i;
400 // Only set the video stream if not album art
401 if (self->video_index < 0 && !(context->streams[i]->disposition & AV_DISPOSITION_ATTACHED_PIC)) {
402 self->video_index = i;
403 }
404 mlt_properties_set( meta_media, key, "video" );
405 snprintf( key, sizeof(key), "meta.media.%u.stream.frame_rate", i );
406 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
407 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
408
409 #if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(56, 1, 0)
410 const char *projection = get_projection(context->streams[i]);
411 if (projection) {
412 snprintf(key, sizeof(key), "meta.media.%u.stream.projection", i);
413 mlt_properties_set_string(meta_media, key, projection);
414 }
415 #endif
416 snprintf( key, sizeof(key), "meta.media.%u.stream.sample_aspect_ratio", i );
417 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
418 snprintf( key, sizeof(key), "meta.media.%u.codec.width", i );
419 mlt_properties_set_int( meta_media, key, codec_params->width );
420 snprintf( key, sizeof(key), "meta.media.%u.codec.height", i );
421 mlt_properties_set_int( meta_media, key, codec_params->height );
422 snprintf( key, sizeof(key), "meta.media.%u.codec.rotate", i );
423 mlt_properties_set_int( meta_media, key, get_rotation(context->streams[i]) );
424 snprintf( key, sizeof(key), "meta.media.%u.codec.frame_rate", i );
425 AVRational frame_rate = { codec_context->time_base.den, codec_context->time_base.num * codec_context->ticks_per_frame };
426 mlt_properties_set_double( meta_media, key, av_q2d( frame_rate ) );
427 snprintf( key, sizeof(key), "meta.media.%u.codec.pix_fmt", i );
428 mlt_properties_set( meta_media, key, av_get_pix_fmt_name( codec_params->format ) );
429 snprintf( key, sizeof(key), "meta.media.%u.codec.sample_aspect_ratio", i );
430 mlt_properties_set_double( meta_media, key, av_q2d( codec_params->sample_aspect_ratio ) );
431 snprintf( key, sizeof(key), "meta.media.%u.codec.colorspace", i );
432 switch ( codec_params->color_space )
433 {
434 case AVCOL_SPC_SMPTE240M:
435 mlt_properties_set_int( meta_media, key, 240 );
436 break;
437 case AVCOL_SPC_BT470BG:
438 case AVCOL_SPC_SMPTE170M:
439 mlt_properties_set_int( meta_media, key, 601 );
440 break;
441 case AVCOL_SPC_BT709:
442 mlt_properties_set_int( meta_media, key, 709 );
443 break;
444 case AVCOL_SPC_UNSPECIFIED:
445 case AVCOL_SPC_RESERVED:
446 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
447 mlt_properties_set_int( meta_media, key, codec_params->width * codec_params->height > 750000 ? 709 : 601 );
448 break;
449 default:
450 mlt_properties_set_int( meta_media, key, codec_context->colorspace );
451 break;
452 }
453 if ( codec_params->color_trc && codec_params->color_trc != AVCOL_TRC_UNSPECIFIED )
454 {
455 snprintf( key, sizeof(key), "meta.media.%u.codec.color_trc", i );
456 mlt_properties_set_double( meta_media, key, codec_params->color_trc );
457 }
458 break;
459 case AVMEDIA_TYPE_AUDIO:
460 if ( !codec_params->channels )
461 break;
462 // Use first audio stream
463 if ( self->audio_index < 0 && pick_audio_format( codec_params->format ) != mlt_audio_none )
464 self->audio_index = i;
465
466 mlt_properties_set( meta_media, key, "audio" );
467 snprintf( key, sizeof(key), "meta.media.%u.codec.sample_fmt", i );
468 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_params->format ) );
469 snprintf( key, sizeof(key), "meta.media.%u.codec.sample_rate", i );
470 mlt_properties_set_int( meta_media, key, codec_params->sample_rate );
471 snprintf( key, sizeof(key), "meta.media.%u.codec.channels", i );
472 mlt_properties_set_int( meta_media, key, codec_params->channels );
473 break;
474 default:
475 break;
476 }
477 // snprintf( key, sizeof(key), "meta.media.%u.stream.time_base", i );
478 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
479 snprintf( key, sizeof(key), "meta.media.%u.codec.name", i );
480 mlt_properties_set( meta_media, key, codec->name );
481 snprintf( key, sizeof(key), "meta.media.%u.codec.long_name", i );
482 mlt_properties_set( meta_media, key, codec->long_name );
483 snprintf( key, sizeof(key), "meta.media.%u.codec.bit_rate", i );
484 mlt_properties_set_int64( meta_media, key, codec_params->bit_rate );
485 // snprintf( key, sizeof(key), "meta.media.%u.codec.time_base", i );
486 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
487 // snprintf( key, sizeof(key), "meta.media.%u.codec.profile", i );
488 // mlt_properties_set_int( meta_media, key, codec_context->profile );
489 // snprintf( key, sizeof(key), "meta.media.%u.codec.level", i );
490 // mlt_properties_set_int( meta_media, key, codec_context->level );
491
492 // Read Metadata
493 while ( ( tag = av_dict_get( stream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
494 {
495 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
496 {
497 snprintf( key, sizeof(key), "meta.attr.%u.stream.%s.markup", i, tag->key );
498 char* value = filter_restricted( tag->value );
499 mlt_properties_set( meta_media, key, value );
500 free( value );
501 }
502 }
503 }
504
505 // Use the album art if that is all we have
506 if (self->video_index < 0 && first_video_index >= 0)
507 self->video_index = first_video_index;
508
509 while ( ( tag = av_dict_get( context->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
510 {
511 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
512 {
513 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
514 char* value = filter_restricted( tag->value );
515 mlt_properties_set( meta_media, key, value );
516 free( value );
517 }
518 }
519
520 return meta_media;
521 }
522
get_aspect_ratio(mlt_properties properties,AVStream * stream,AVCodecParameters * codec_params)523 static void get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecParameters *codec_params )
524 {
525 AVRational sar = stream->sample_aspect_ratio;
526 if ( sar.num <= 0 || sar.den <= 0 )
527 sar = codec_params->sample_aspect_ratio;
528 if ( sar.num <= 0 || sar.den <= 0 )
529 sar.num = sar.den = 1;
530 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", sar.num );
531 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", sar.den );
532 mlt_properties_set_double( properties, "aspect_ratio", av_q2d( sar ) );
533 }
534
parse_url(mlt_profile profile,const char * URL,AVInputFormat ** format,AVDictionary ** params)535 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVDictionary **params )
536 {
537 (void) profile; // unused
538 if ( !URL ) return NULL;
539
540 char *protocol = strdup( URL );
541 char *url = strchr( protocol, ':' );
542
543 // Truncate protocol string
544 if (url && (url - protocol) > 1 && avio_check(URL, 0) < 0) { // if defined and not a drive letter
545 url[0] = '\0';
546 ++url;
547 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url );
548
549 // Lookup the format
550 *format = av_find_input_format( protocol );
551 } else {
552 url = protocol;
553 }
554
555 // Eat the format designator
556 char *result = url;
557
558 // support for legacy width and height parameters
559 char *width = NULL;
560 char *height = NULL;
561
562 // Parse out params
563 char* query = strchr( url, '?' );
564 if (*format) {
565 // Query string delimiter is '?'
566 url = ( query && query > url && query[-1] != '\\' ) ? query : NULL;
567 } else {
568 // Ignore unescaped question marks
569 while ( query && query > url && query[-1] != '\\' ) {
570 query = strchr( query + 1, '?' );
571 }
572 // Query string delimiter is '\?'
573 url = ( query && query > url && query[-1] == '\\' ) ? query : NULL;
574 if (url) url[-1] = '\0'; // null the backslash
575 }
576 while ( url )
577 {
578 url[0] = '\0';
579 char *name = strdup( ++url );
580 char *value = strchr( name, '=' );
581 if ( !value )
582 // Also accept : as delimiter for backwards compatibility.
583 value = strchr( name, ':' );
584 if ( value )
585 {
586 value[0] = '\0';
587 value++;
588 char *t = strchr( value, '&' );
589 if ( t )
590 t[0] = 0;
591 // translate old parameters to new av_dict names
592 if ( !strcmp( name, "frame_rate" ) )
593 av_dict_set( params, "framerate", value, 0 );
594 else if ( !strcmp( name, "pix_fmt" ) )
595 av_dict_set( params, "pixel_format", value, 0 );
596 else if ( !strcmp( name, "width" ) )
597 width = strdup( value );
598 else if ( !strcmp( name, "height" ) )
599 height = strdup( value );
600 else
601 // generic demux/device option support
602 av_dict_set( params, name, value, 0 );
603 }
604 free( name );
605 url = strchr( url, '&' );
606 }
607 // continued support for legacy width and height parameters
608 if ( width && height )
609 {
610 char *s = malloc( strlen( width ) + strlen( height ) + 2 );
611 strcpy( s, width );
612 strcat( s, "x");
613 strcat( s, height );
614 av_dict_set( params, "video_size", s, 0 );
615 free( s );
616 }
617 free( width );
618 free( height );
619
620 result = strdup(result);
621 free( protocol );
622 mlt_log_debug(NULL, "[producer avformat] %s filename = %s\n", __FUNCTION__, result);
623 return result;
624 }
625
pick_pix_fmt(enum AVPixelFormat pix_fmt)626 static enum AVPixelFormat pick_pix_fmt( enum AVPixelFormat pix_fmt )
627 {
628 switch ( pix_fmt )
629 {
630 case AV_PIX_FMT_ARGB:
631 case AV_PIX_FMT_RGBA:
632 case AV_PIX_FMT_ABGR:
633 case AV_PIX_FMT_BGRA:
634 return AV_PIX_FMT_RGBA;
635 #if defined(FFUDIV)
636 case AV_PIX_FMT_BAYER_RGGB16LE:
637 return AV_PIX_FMT_RGB24;
638 #endif
639 #if USE_HWACCEL
640 case AV_PIX_FMT_VAAPI:
641 case AV_PIX_FMT_CUDA:
642 case AV_PIX_FMT_VIDEOTOOLBOX:
643 case AV_PIX_FMT_DXVA2_VLD:
644 case AV_PIX_FMT_D3D11:
645 return AV_PIX_FMT_YUV420P;
646 #endif
647 default:
648 return AV_PIX_FMT_YUV422P;
649 }
650 }
651
get_basic_info(producer_avformat self,mlt_profile profile,const char * filename)652 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
653 {
654 int error = 0;
655
656 // Get the properties
657 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
658
659 AVFormatContext *format = self->video_format;
660
661 // Get the duration
662 if ( mlt_properties_get_position( properties, "length" ) <= 0 ||
663 mlt_properties_get_position( properties, "out" ) <= 0 )
664 {
665 if ( format->duration != AV_NOPTS_VALUE )
666 {
667 // This isn't going to be accurate for all formats
668 // We will treat everything with the producer fps.
669 mlt_position frames = ( mlt_position ) lrint( format->duration * mlt_profile_fps( profile ) / AV_TIME_BASE );
670 if ( mlt_properties_get_position( properties, "out" ) <= 0 )
671 mlt_properties_set_position( properties, "out", frames - 1 );
672 if ( mlt_properties_get_position( properties, "length" ) <= 0 )
673 mlt_properties_set_position( properties, "length", frames );
674 }
675 else if ( format->nb_streams > 0 && format->streams[0]->codecpar && format->streams[0]->codecpar->codec_id == AV_CODEC_ID_WEBP )
676 {
677 char *e = getenv( "MLT_DEFAULT_PRODUCER_LENGTH" );
678 int p = e ? atoi( e ) : 15000;
679 mlt_properties_set_int( properties, "out", MAX(0, p - 1) );
680 mlt_properties_set_int( properties, "length", p );
681 }
682 else
683 {
684 // Set live sources to run forever
685 if ( mlt_properties_get_position( properties, "length" ) <= 0 )
686 mlt_properties_set_position( properties, "length", INT_MAX );
687 if ( mlt_properties_get_position( properties, "out" ) <= 0 )
688 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
689 mlt_properties_set( properties, "eof", "loop" );
690 }
691 }
692
693 // Check if we're seekable
694 // avdevices are typically AVFMT_NOFILE and not seekable
695 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
696 if ( format->pb )
697 {
698 // protocols can indicate if they support seeking
699 self->seekable = format->pb->seekable;
700 }
701 if ( self->seekable )
702 {
703 // Do a more rigorous test of seekable on a disposable context
704 if ( format->nb_streams > 0 && format->streams[0]->codecpar && format->streams[0]->codecpar->codec_id != AV_CODEC_ID_WEBP )
705 self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
706 mlt_properties_set_int( properties, "seekable", self->seekable );
707 self->dummy_context = format;
708 self->video_format = NULL;
709 avformat_open_input( &self->video_format, filename, NULL, NULL );
710 avformat_find_stream_info( self->video_format, NULL );
711 format = self->video_format;
712 }
713 self->video_seekable = self->seekable;
714
715 // Fetch the width, height and aspect ratio
716 if ( self->video_index != -1 )
717 {
718 AVCodecParameters *codec_params = format->streams[ self->video_index ]->codecpar;
719 mlt_properties_set_int( properties, "width", codec_params->width );
720 mlt_properties_set_int( properties, "height", codec_params->height );
721 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_params );
722
723 int pix_fmt = codec_params->format;
724 pick_av_pixel_format( &pix_fmt );
725 if ( pix_fmt != AV_PIX_FMT_NONE ) {
726 // Verify that we can convert this to one of our image formats.
727 struct SwsContext *context = sws_getContext( codec_params->width, codec_params->height, pix_fmt,
728 codec_params->width, codec_params->height, pick_pix_fmt( codec_params->format ), SWS_BILINEAR, NULL, NULL, NULL);
729 if ( context )
730 sws_freeContext( context );
731 else
732 error = 1;
733 } else {
734 self->video_index = -1;
735 }
736 }
737 return error;
738 }
739
740 #ifdef AVFILTER
setup_video_filters(producer_avformat self)741 static int setup_video_filters( producer_avformat self )
742 {
743 mlt_properties properties = MLT_PRODUCER_PROPERTIES(self->parent);
744 AVFormatContext *format = self->video_format;
745 AVStream* stream = format->streams[ self->video_index ];
746 AVCodecParameters *codec_params = stream->codecpar;
747
748 self->vfilter_graph = avfilter_graph_alloc();
749
750 // From ffplay.c:configure_video_filters().
751 char buffersrc_args[256];
752 snprintf(buffersrc_args, sizeof(buffersrc_args),
753 "video_size=%dx%d:pix_fmt=%d:time_base=%d/%d:pixel_aspect=%d/%d:frame_rate=%d/%d",
754 codec_params->width, codec_params->height, codec_params->format,
755 stream->time_base.num, stream->time_base.den,
756 mlt_properties_get_int(properties, "meta.media.sample_aspect_num"),
757 FFMAX(mlt_properties_get_int(properties, "meta.media.sample_aspect_den"), 1),
758 stream->avg_frame_rate.num, FFMAX(stream->avg_frame_rate.den, 1));
759
760 int result = avfilter_graph_create_filter(&self->vfilter_in, avfilter_get_by_name("buffer"),
761 "mlt_buffer", buffersrc_args, NULL, self->vfilter_graph);
762
763 if (result >= 0) {
764 result = avfilter_graph_create_filter(&self->vfilter_out, avfilter_get_by_name("buffersink"),
765 "mlt_buffersink", NULL, NULL, self->vfilter_graph);
766
767 if (result >= 0) {
768 enum AVPixelFormat pix_fmts[] = { codec_params->format, AV_PIX_FMT_NONE };
769 result = av_opt_set_int_list(self->vfilter_out, "pix_fmts", pix_fmts, AV_PIX_FMT_NONE, AV_OPT_SEARCH_CHILDREN);
770 }
771 }
772
773 return result;
774 }
775
insert_filter(AVFilterGraph * graph,AVFilterContext ** last_filter,const char * name,const char * args)776 static int insert_filter(AVFilterGraph *graph, AVFilterContext **last_filter, const char *name, const char *args)
777 {
778 AVFilterContext *filt_ctx;
779 int result = avfilter_graph_create_filter(&filt_ctx, avfilter_get_by_name(name),
780 name, args, NULL, graph);
781 if (result >= 0) {
782 result = avfilter_link(filt_ctx, 0, *last_filter, 0);
783 if (result >= 0)
784 *last_filter = filt_ctx;
785 }
786 return result;
787 }
788 #endif
789
790 /** Open the file.
791 */
792
producer_open(producer_avformat self,mlt_profile profile,const char * URL,int take_lock,int test_open)793 static int producer_open(producer_avformat self, mlt_profile profile, const char *URL, int take_lock, int test_open )
794 {
795 // Return an error code (0 == no error)
796 int error = 0;
797 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
798
799 if ( !self->is_mutex_init )
800 {
801 pthread_mutexattr_t attr;
802 pthread_mutexattr_init(&attr);
803 pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
804 pthread_mutex_init( &self->audio_mutex, &attr );
805 pthread_mutex_init( &self->video_mutex, &attr );
806 pthread_mutex_init( &self->packets_mutex, &attr );
807 pthread_mutex_init( &self->open_mutex, &attr );
808 self->is_mutex_init = 1;
809 }
810
811 // Lock the service
812 if ( take_lock )
813 {
814 pthread_mutex_lock( &self->audio_mutex );
815 pthread_mutex_lock( &self->video_mutex );
816 }
817 mlt_events_block( properties, self->parent );
818
819 // Parse URL
820 AVInputFormat *format = NULL;
821 AVDictionary *params = NULL;
822 char *filename = parse_url( profile, URL, &format, ¶ms );
823
824 // Now attempt to open the file or device with filename
825 error = avformat_open_input( &self->video_format, filename, format, ¶ms ) < 0;
826 if ( error )
827 // If the URL is a network stream URL, then we probably need to open with full URL
828 error = avformat_open_input( &self->video_format, URL, format, ¶ms ) < 0;
829
830 // Set MLT properties onto video AVFormatContext
831 if ( !error && self->video_format )
832 {
833 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
834 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
835 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
836 }
837
838 // If successful, then try to get additional info
839 if ( !error && self->video_format )
840 {
841 // Get the stream info
842 error = avformat_find_stream_info( self->video_format, NULL ) < 0;
843
844 // Continue if no error
845 if ( !error && self->video_format )
846 {
847 // Find default audio and video streams
848 find_default_streams( self );
849 error = get_basic_info( self, profile, filename );
850
851 // Initialize position info
852 self->first_pts = AV_NOPTS_VALUE;
853 self->last_position = POSITION_INITIAL;
854
855 #if USE_HWACCEL
856 AVDictionaryEntry *hwaccel = av_dict_get( params, "hwaccel", NULL, 0 );
857 AVDictionaryEntry *hwaccel_device = av_dict_get( params, "hwaccel_device", NULL, 0 );
858
859 if ( hwaccel && hwaccel->value )
860 {
861 // Leaving `device=NULL` will cause query string parameter `hwaccel_device` to be ignored
862 char *device = NULL;
863 if ( !strcmp( hwaccel->value, "vaapi" ) )
864 {
865 self->hwaccel.pix_fmt = AV_PIX_FMT_VAAPI;
866 self->hwaccel.device_type = AV_HWDEVICE_TYPE_VAAPI;
867 device = "/dev/dri/renderD128";
868 }
869 else if ( !strcmp( hwaccel->value, "cuda" ) )
870 {
871 self->hwaccel.pix_fmt = AV_PIX_FMT_CUDA;
872 self->hwaccel.device_type = AV_HWDEVICE_TYPE_CUDA;
873 device = "0";
874 }
875 else if ( !strcmp( hwaccel->value, "videotoolbox" ) )
876 {
877 self->hwaccel.pix_fmt = AV_PIX_FMT_VIDEOTOOLBOX;
878 self->hwaccel.device_type = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
879 }
880 else if ( !strcmp( hwaccel->value, "d3d11va" ) )
881 {
882 self->hwaccel.pix_fmt = AV_PIX_FMT_D3D11;
883 self->hwaccel.device_type = AV_HWDEVICE_TYPE_D3D11VA;
884 device = "0";
885 }
886 else if ( !strcmp( hwaccel->value, "dxva2" ) )
887 {
888 self->hwaccel.pix_fmt = AV_PIX_FMT_DXVA2_VLD;
889 self->hwaccel.device_type = AV_HWDEVICE_TYPE_DXVA2;
890 device = "0";
891 }
892 else
893 {
894 // TODO: init other hardware types
895 }
896
897 if (device) {
898 if (hwaccel_device && hwaccel_device->value)
899 device = hwaccel_device->value;
900 memcpy(self->hwaccel.device, device, strlen(device));
901 }
902 }
903 #endif
904
905 if ( !self->audio_format )
906 {
907 // We're going to cheat here - for seekable A/V files, we will have separate contexts
908 // to support independent seeking of audio from video.
909 // TODO: Is this really necessary?
910 if ( self->audio_index != -1 && self->video_index != -1 )
911 {
912 if ( self->seekable )
913 {
914 // And open again for our audio context
915 avformat_open_input( &self->audio_format, filename, NULL, NULL );
916 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
917 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
918 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
919 avformat_find_stream_info( self->audio_format, NULL );
920 }
921 else
922 {
923 self->audio_format = self->video_format;
924 }
925 }
926 else if ( self->audio_index != -1 )
927 {
928 // We only have an audio context
929 self->audio_format = self->video_format;
930 self->video_format = NULL;
931 }
932 else if ( self->video_index == -1 )
933 {
934 // Something has gone wrong
935 error = -1;
936 }
937 if ( self->audio_format && !self->audio_streams )
938 get_audio_streams_info( self );
939
940 #ifdef AVFILTER
941 // Setup autorotate filters.
942 if (self->video_index != -1) {
943 self->autorotate = !mlt_properties_get(properties, "autorotate") || mlt_properties_get_int(properties, "autorotate");
944 if (!test_open && self->autorotate && !self->vfilter_graph) {
945 double theta = get_rotation(self->video_format->streams[self->video_index]);
946
947 if (fabs(theta - 90) < 1.0) {
948 error = ( setup_video_filters(self) < 0 );
949 AVFilterContext *last_filter = self->vfilter_out;
950 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "transpose", "clock") < 0 );
951 if (!error) error = ( avfilter_link(self->vfilter_in, 0, last_filter, 0) < 0 );
952 if (!error) error = ( avfilter_graph_config(self->vfilter_graph, NULL) < 0 );
953 } else if (fabs(theta - 180) < 1.0) {
954 error = ( setup_video_filters(self) < 0 );
955 AVFilterContext *last_filter = self->vfilter_out;
956 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "hflip", NULL) < 0 );
957 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "vflip", NULL) < 0 );
958 if (!error) error = ( avfilter_link(self->vfilter_in, 0, last_filter, 0) < 0 );
959 if (!error) error = ( avfilter_graph_config(self->vfilter_graph, NULL) < 0 );
960 } else if (fabs(theta - 270) < 1.0) {
961 error = ( setup_video_filters(self) < 0 );
962 AVFilterContext *last_filter = self->vfilter_out;
963 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "transpose", "cclock") < 0 );
964 if (!error) error = ( avfilter_link(self->vfilter_in, 0, last_filter, 0) < 0 );
965 if (!error) error = ( avfilter_graph_config(self->vfilter_graph, NULL) < 0 );
966 }
967 }
968 }
969 #endif
970 }
971 }
972 }
973 av_dict_free( ¶ms );
974 free( filename );
975 if ( !error )
976 {
977 self->apackets = mlt_deque_init();
978 self->vpackets = mlt_deque_init();
979 }
980
981 if ( self->dummy_context )
982 {
983 pthread_mutex_lock( &self->open_mutex );
984 avformat_close_input( &self->dummy_context );
985 self->dummy_context = NULL;
986 pthread_mutex_unlock( &self->open_mutex );
987 }
988
989 // Unlock the service
990 if ( take_lock )
991 {
992 pthread_mutex_unlock( &self->audio_mutex );
993 pthread_mutex_unlock( &self->video_mutex );
994 }
995 mlt_events_unblock( properties, self->parent );
996
997 return error;
998 }
999
prepare_reopen(producer_avformat self)1000 static void prepare_reopen( producer_avformat self )
1001 {
1002 mlt_service_lock( MLT_PRODUCER_SERVICE( self->parent ) );
1003 pthread_mutex_lock( &self->audio_mutex );
1004 pthread_mutex_lock( &self->open_mutex );
1005
1006 int i;
1007 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
1008 {
1009 mlt_pool_release( self->audio_buffer[i] );
1010 self->audio_buffer[i] = NULL;
1011 av_free( self->decode_buffer[i] );
1012 self->decode_buffer[i] = NULL;
1013 if ( self->audio_codec[i] )
1014 avcodec_close( self->audio_codec[i] );
1015 self->audio_codec[i] = NULL;
1016 }
1017 if ( self->video_codec )
1018 avcodec_close( self->video_codec );
1019 self->video_codec = NULL;
1020 av_frame_unref( self->video_frame );
1021 #if USE_HWACCEL
1022 av_buffer_unref( &self->hwaccel.device_ctx );
1023 self->hwaccel.device_ctx = NULL;
1024 #endif
1025 if ( self->seekable && self->audio_format )
1026 avformat_close_input( &self->audio_format );
1027 if ( self->video_format )
1028 avformat_close_input( &self->video_format );
1029 self->audio_format = NULL;
1030 self->video_format = NULL;
1031 #ifdef AVFILTER
1032 avfilter_graph_free( &self->vfilter_graph );
1033 #endif
1034 pthread_mutex_unlock( &self->open_mutex );
1035
1036 // Cleanup the packet queues
1037 AVPacket *pkt;
1038 if ( self->apackets )
1039 {
1040 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
1041 {
1042 av_packet_unref( pkt );
1043 free( pkt );
1044 }
1045 mlt_deque_close( self->apackets );
1046 self->apackets = NULL;
1047 }
1048 if ( self->vpackets )
1049 {
1050 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
1051 {
1052 av_packet_unref( pkt );
1053 free( pkt );
1054 }
1055 mlt_deque_close( self->vpackets );
1056 self->vpackets = NULL;
1057 }
1058 pthread_mutex_unlock( &self->audio_mutex );
1059 mlt_service_unlock( MLT_PRODUCER_SERVICE( self->parent ) );
1060 }
1061
best_pts(producer_avformat self,int64_t pts,int64_t dts)1062 static int64_t best_pts( producer_avformat self, int64_t pts, int64_t dts )
1063 {
1064 self->invalid_pts_counter += pts == AV_NOPTS_VALUE;
1065 self->invalid_dts_counter += dts == AV_NOPTS_VALUE;
1066 if ( ( self->invalid_pts_counter <= self->invalid_dts_counter
1067 || dts == AV_NOPTS_VALUE ) && pts != AV_NOPTS_VALUE )
1068 return pts;
1069 else
1070 return dts;
1071 }
1072
find_first_pts(producer_avformat self,int video_index)1073 static void find_first_pts( producer_avformat self, int video_index )
1074 {
1075 // find initial PTS
1076 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
1077 int ret = 0;
1078 int pkt_countdown = 500; // check max 500 packets for first video keyframe PTS
1079 int vfr_countdown = 20; // check max 20 video frames for VFR
1080 int vfr_counter = 0; // counts the number of frame duration changes
1081 AVPacket pkt;
1082 int64_t prev_pkt_duration = AV_NOPTS_VALUE;
1083
1084 av_init_packet( &pkt );
1085 while ( ret >= 0 && pkt_countdown-- > 0 &&
1086 ( self->first_pts == AV_NOPTS_VALUE || ( vfr_counter < VFR_THRESHOLD && vfr_countdown > 0 ) ) )
1087 {
1088 ret = av_read_frame( context, &pkt );
1089 if ( ret >= 0 && pkt.stream_index == video_index )
1090 {
1091 // Variable frame rate check
1092 if ( pkt.duration != AV_NOPTS_VALUE && pkt.duration != prev_pkt_duration ) {
1093 mlt_log_verbose( MLT_PRODUCER_SERVICE(self->parent), "checking VFR: pkt.duration %"PRId64"\n", pkt.duration );
1094 if ( prev_pkt_duration != AV_NOPTS_VALUE )
1095 ++vfr_counter;
1096 }
1097 prev_pkt_duration = pkt.duration;
1098 vfr_countdown--;
1099
1100 // Finding PTS of first video key frame
1101 if ( ( pkt.flags & AV_PKT_FLAG_KEY ) && self->first_pts == AV_NOPTS_VALUE )
1102 {
1103 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
1104 "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n",
1105 pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
1106 if ( pkt.dts != AV_NOPTS_VALUE && pkt.dts < 0 )
1107 // Decoding Time Stamps with negative values are reported by ffmpeg code for
1108 // (at least) MP4 files containing h.264 video using b-frames.
1109 // For reasons not understood yet, the first PTS computed then is that of the
1110 // third frame, causing MLT to display the third frame as if it was the first.
1111 // This if-clause is meant to catch and work around this issue - if there is
1112 // a valid, but negative DTS value, we just guess that the first valid
1113 // Presentation Time Stamp is == 0.
1114 self->first_pts = 0;
1115 else
1116 self->first_pts = best_pts( self, pkt.pts, pkt.dts );
1117 }
1118 }
1119 av_packet_unref( &pkt );
1120 }
1121 if ( vfr_counter >= VFR_THRESHOLD )
1122 mlt_properties_set_int( MLT_PRODUCER_PROPERTIES(self->parent), "meta.media.variable_frame_rate", 1 );
1123 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
1124 }
1125
seek_video(producer_avformat self,mlt_position position,int64_t req_position,int preseek)1126 static int seek_video( producer_avformat self, mlt_position position,
1127 int64_t req_position, int preseek )
1128 {
1129 mlt_producer producer = self->parent;
1130 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1131 int paused = 0;
1132 int seek_threshold = mlt_properties_get_int( properties, "seek_threshold" );
1133 if ( seek_threshold <= 0 ) seek_threshold = 12;
1134
1135 pthread_mutex_lock( &self->packets_mutex );
1136
1137 if ( self->video_seekable && ( position != self->video_expected || self->last_position < 0 ) )
1138 {
1139
1140 // Fetch the video format context
1141 AVFormatContext *context = self->video_format;
1142
1143 // Get the video stream
1144 AVStream *stream = context->streams[ self->video_index ];
1145
1146 // Get codec context
1147 AVCodecContext *codec_context = stream->codec;
1148
1149 // We may want to use the source fps if available
1150 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1151 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1152
1153 if ( self->first_pts == AV_NOPTS_VALUE && self->last_position == POSITION_INITIAL )
1154 find_first_pts( self, self->video_index );
1155
1156 if ( self->video_frame && position + 1 == self->video_expected )
1157 {
1158 // We're paused - use last image
1159 paused = 1;
1160 }
1161 else if ( position < self->video_expected || position - self->video_expected >= seek_threshold || self->last_position < 0 )
1162 {
1163 // Calculate the timestamp for the requested frame
1164 int64_t timestamp = req_position / ( av_q2d( self->video_time_base ) * source_fps );
1165 if ( req_position <= 0 )
1166 timestamp = 0;
1167 else if ( self->first_pts != AV_NOPTS_VALUE )
1168 timestamp += self->first_pts;
1169 else if ( context->start_time != AV_NOPTS_VALUE )
1170 timestamp += context->start_time;
1171 if ( preseek && av_q2d( self->video_time_base ) != 0 )
1172 timestamp -= 2 / av_q2d( self->video_time_base );
1173 if ( timestamp < 0 )
1174 timestamp = 0;
1175 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position " MLT_POSITION_FMT " expected "MLT_POSITION_FMT" last_pos %"PRId64"\n",
1176 timestamp, position, self->video_expected, self->last_position );
1177
1178 // Seek to the timestamp
1179 codec_context->skip_loop_filter = AVDISCARD_NONREF;
1180 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
1181
1182 // flush any pictures still in decode buffer
1183 avcodec_flush_buffers( codec_context );
1184 self->video_send_result = 0;
1185
1186 // Remove the cached info relating to the previous position
1187 self->current_position = POSITION_INVALID;
1188 self->last_position = POSITION_INVALID;
1189 av_frame_unref(self->video_frame);
1190 }
1191 }
1192 pthread_mutex_unlock( &self->packets_mutex );
1193 return paused;
1194 }
1195
1196 /** Convert a frame position to a time code.
1197 */
1198
producer_time_of_frame(mlt_producer producer,mlt_position position)1199 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
1200 {
1201 return ( double )position / mlt_producer_get_fps( producer );
1202 }
1203
1204 // Collect information about all audio streams
1205
get_audio_streams_info(producer_avformat self)1206 static void get_audio_streams_info( producer_avformat self )
1207 {
1208 // Fetch the audio format context
1209 AVFormatContext *context = self->audio_format;
1210 unsigned int i;
1211
1212 for ( i = 0;
1213 i < context->nb_streams;
1214 i++ )
1215 {
1216 if ( context->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO )
1217 {
1218 AVCodecParameters *codec_params = context->streams[i]->codecpar;
1219 AVCodec *codec = avcodec_find_decoder( codec_params->codec_id );
1220
1221 // If we don't have a codec and we can't initialise it, we can't do much more...
1222 pthread_mutex_lock( &self->open_mutex );
1223 if ( codec && avcodec_open2( context->streams[i]->codec, codec, NULL ) >= 0 )
1224 {
1225 self->audio_streams++;
1226 self->audio_max_stream = i;
1227 self->total_channels += codec_params->channels;
1228 if ( codec_params->channels > self->max_channel )
1229 self->max_channel = codec_params->channels;
1230 if ( codec_params->sample_rate > self->max_frequency )
1231 self->max_frequency = codec_params->sample_rate;
1232 avcodec_close( context->streams[i]->codec );
1233 }
1234 pthread_mutex_unlock( &self->open_mutex );
1235 }
1236 }
1237 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1238 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1239 }
1240
pick_image_format(enum AVPixelFormat pix_fmt)1241 static mlt_image_format pick_image_format( enum AVPixelFormat pix_fmt )
1242 {
1243 switch ( pix_fmt )
1244 {
1245 case AV_PIX_FMT_ARGB:
1246 case AV_PIX_FMT_RGBA:
1247 case AV_PIX_FMT_ABGR:
1248 case AV_PIX_FMT_BGRA:
1249 return mlt_image_rgba;
1250 case AV_PIX_FMT_YUV420P:
1251 case AV_PIX_FMT_YUVJ420P:
1252 case AV_PIX_FMT_YUVA420P:
1253 return mlt_image_yuv420p;
1254 case AV_PIX_FMT_RGB24:
1255 case AV_PIX_FMT_BGR24:
1256 case AV_PIX_FMT_GRAY8:
1257 case AV_PIX_FMT_MONOWHITE:
1258 case AV_PIX_FMT_MONOBLACK:
1259 case AV_PIX_FMT_RGB8:
1260 case AV_PIX_FMT_BGR8:
1261 #if defined(FFUDIV)
1262 case AV_PIX_FMT_BAYER_RGGB16LE:
1263 return mlt_image_rgb;
1264 #endif
1265 default:
1266 return mlt_image_yuv422;
1267 }
1268 }
1269
pick_audio_format(int sample_fmt)1270 static mlt_audio_format pick_audio_format( int sample_fmt )
1271 {
1272 switch ( sample_fmt )
1273 {
1274 // interleaved
1275 case AV_SAMPLE_FMT_U8:
1276 return mlt_audio_u8;
1277 case AV_SAMPLE_FMT_S16:
1278 return mlt_audio_s16;
1279 case AV_SAMPLE_FMT_S32:
1280 return mlt_audio_s32le;
1281 case AV_SAMPLE_FMT_FLT:
1282 return mlt_audio_f32le;
1283 // planar - this producer converts planar to interleaved
1284 case AV_SAMPLE_FMT_U8P:
1285 return mlt_audio_u8;
1286 case AV_SAMPLE_FMT_S16P:
1287 return mlt_audio_s16;
1288 case AV_SAMPLE_FMT_S32P:
1289 return mlt_audio_s32le;
1290 case AV_SAMPLE_FMT_FLTP:
1291 return mlt_audio_f32le;
1292 default:
1293 return mlt_audio_none;
1294 }
1295 }
1296
1297 /**
1298 * Handle deprecated pixel format (JPEG range in YUV420P for example).
1299 *
1300 * Replace pix_fmt with the official pixel format to use.
1301 * @return 0 if no pix_fmt replacement, 1 otherwise
1302 */
pick_av_pixel_format(int * pix_fmt)1303 static int pick_av_pixel_format( int *pix_fmt )
1304 {
1305 #if defined(FFUDIV)
1306 switch (*pix_fmt)
1307 {
1308 case AV_PIX_FMT_YUVJ420P:
1309 *pix_fmt = AV_PIX_FMT_YUV420P;
1310 return 1;
1311 case AV_PIX_FMT_YUVJ411P:
1312 *pix_fmt = AV_PIX_FMT_YUV411P;
1313 return 1;
1314 case AV_PIX_FMT_YUVJ422P:
1315 *pix_fmt = AV_PIX_FMT_YUV422P;
1316 return 1;
1317 case AV_PIX_FMT_YUVJ444P:
1318 *pix_fmt = AV_PIX_FMT_YUV444P;
1319 return 1;
1320 case AV_PIX_FMT_YUVJ440P:
1321 *pix_fmt = AV_PIX_FMT_YUV440P;
1322 return 1;
1323 }
1324 #endif
1325 return 0;
1326 }
1327
1328 #if defined(FFUDIV) && (LIBSWSCALE_VERSION_INT >= ((3<<16)+(1<<8)+101))
1329 struct sliced_pix_fmt_conv_t
1330 {
1331 int width, height, slice_w;
1332 AVFrame *frame;
1333 uint8_t *out_data[4];
1334 int out_stride[4];
1335 enum AVPixelFormat src_format, dst_format;
1336 const AVPixFmtDescriptor *src_desc, *dst_desc;
1337 int flags, src_colorspace, dst_colorspace, src_full_range, dst_full_range;
1338 };
1339
sliced_h_pix_fmt_conv_proc(int id,int idx,int jobs,void * cookie)1340 static int sliced_h_pix_fmt_conv_proc( int id, int idx, int jobs, void* cookie )
1341 {
1342 uint8_t *out[4];
1343 const uint8_t *in[4];
1344 int in_stride[4], out_stride[4];
1345 int src_v_chr_pos = -513, dst_v_chr_pos = -513, ret, i, slice_x, slice_w, h, mul, field, slices, interlaced = 0;
1346
1347 struct SwsContext *sws;
1348 struct sliced_pix_fmt_conv_t* ctx = ( struct sliced_pix_fmt_conv_t* )cookie;
1349
1350 interlaced = ctx->frame->interlaced_frame;
1351 field = ( interlaced ) ? ( idx & 1 ) : 0;
1352 idx = ( interlaced ) ? ( idx / 2 ) : idx;
1353 slices = ( interlaced ) ? ( jobs / 2 ) : jobs;
1354 mul = ( interlaced ) ? 2 : 1;
1355 h = ctx->height >> !!interlaced;
1356 slice_w = ctx->slice_w;
1357 slice_x = slice_w * idx;
1358 slice_w = FFMIN( slice_w, ctx->width - slice_x );
1359
1360 if ( AV_PIX_FMT_YUV420P == ctx->src_format )
1361 src_v_chr_pos = ( !interlaced ) ? 128 : ( !field ) ? 64 : 192;
1362
1363 if ( AV_PIX_FMT_YUV420P == ctx->dst_format )
1364 dst_v_chr_pos = ( !interlaced ) ? 128 : ( !field ) ? 64 : 192;
1365
1366 mlt_log_debug( NULL, "%s:%d: [id=%d, idx=%d, jobs=%d], interlaced=%d, field=%d, slices=%d, mul=%d, h=%d, slice_w=%d, slice_x=%d ctx->src_desc=[log2_chroma_h=%d, log2_chroma_w=%d], src_v_chr_pos=%d, dst_v_chr_pos=%d\n",
1367 __FUNCTION__, __LINE__, id, idx, jobs, interlaced, field, slices, mul, h, slice_w, slice_x, ctx->src_desc->log2_chroma_h, ctx->src_desc->log2_chroma_w, src_v_chr_pos, dst_v_chr_pos );
1368
1369 if ( slice_w <= 0 )
1370 return 0;
1371
1372 sws = sws_alloc_context();
1373
1374 av_opt_set_int( sws, "srcw", slice_w, 0 );
1375 av_opt_set_int( sws, "srch", h, 0 );
1376 av_opt_set_int( sws, "src_format", ctx->src_format, 0 );
1377 av_opt_set_int( sws, "dstw", slice_w, 0 );
1378 av_opt_set_int( sws, "dsth", h, 0 );
1379 av_opt_set_int( sws, "dst_format", ctx->dst_format, 0 );
1380 av_opt_set_int( sws, "sws_flags", ctx->flags, 0 );
1381
1382 av_opt_set_int( sws, "src_h_chr_pos", -513, 0 );
1383 av_opt_set_int( sws, "src_v_chr_pos", src_v_chr_pos, 0 );
1384 av_opt_set_int( sws, "dst_h_chr_pos", -513, 0 );
1385 av_opt_set_int( sws, "dst_v_chr_pos", dst_v_chr_pos, 0 );
1386
1387 if ( ( ret = sws_init_context( sws, NULL, NULL ) ) < 0 )
1388 {
1389 mlt_log_error( NULL, "%s:%d: sws_init_context failed, ret=%d\n", __FUNCTION__, __LINE__, ret );
1390 sws_freeContext( sws );
1391 return 0;
1392 }
1393
1394 mlt_set_luma_transfer( sws, ctx->src_colorspace, ctx->dst_colorspace, ctx->src_full_range, ctx->dst_full_range );
1395
1396 #if LIBAVUTIL_VERSION_INT < AV_VERSION_INT(55, 0, 100)
1397 #define PIX_DESC_BPP(DESC) (DESC.step_minus1 + 1)
1398 #else
1399 #define PIX_DESC_BPP(DESC) (DESC.step)
1400 #endif
1401
1402 for( i = 0; i < 4; i++ )
1403 {
1404 int in_offset = (AV_PIX_FMT_FLAG_PLANAR & ctx->src_desc->flags)
1405 ? ( ( 1 == i || 2 == i ) ? ( slice_x >> ctx->src_desc->log2_chroma_w ) : slice_x )
1406 : ( ( 0 == i ) ? slice_x : 0 );
1407
1408 int out_offset = (AV_PIX_FMT_FLAG_PLANAR & ctx->dst_desc->flags)
1409 ? ( ( 1 == i || 2 == i ) ? ( slice_x >> ctx->dst_desc->log2_chroma_w ) : slice_x )
1410 : ( ( 0 == i ) ? slice_x : 0 );
1411
1412 in_offset *= PIX_DESC_BPP(ctx->src_desc->comp[i]);
1413 out_offset *= PIX_DESC_BPP(ctx->dst_desc->comp[i]);
1414
1415 in_stride[i] = ctx->frame->linesize[i] * mul;
1416 out_stride[i] = ctx->out_stride[i] * mul;
1417
1418 in[i] = ctx->frame->data[i] + ctx->frame->linesize[i] * field + in_offset;
1419 out[i] = ctx->out_data[i] + ctx->out_stride[i] * field + out_offset;
1420 }
1421
1422 sws_scale( sws, in, in_stride, 0, h, out, out_stride );
1423
1424 sws_freeContext( sws );
1425
1426 return 0;
1427 }
1428 #endif
1429
1430 // returns resulting YUV colorspace
convert_image(producer_avformat self,AVFrame * frame,uint8_t * buffer,int pix_fmt,mlt_image_format * format,int width,int height,uint8_t ** alpha)1431 static int convert_image( producer_avformat self, AVFrame *frame, uint8_t *buffer, int pix_fmt,
1432 mlt_image_format *format, int width, int height, uint8_t **alpha )
1433 {
1434 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
1435 int result = self->yuv_colorspace;
1436
1437 mlt_log_timings_begin();
1438
1439 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "%s @ %dx%d space %d->%d\n",
1440 mlt_image_format_name( *format ),
1441 width, height, self->yuv_colorspace, profile->colorspace );
1442
1443 // extract alpha from planar formats
1444 if ( ( pix_fmt == AV_PIX_FMT_YUVA420P
1445 #if defined(FFUDIV)
1446 || pix_fmt == AV_PIX_FMT_YUVA444P
1447 #endif
1448 ) &&
1449 *format != mlt_image_rgba &&
1450 frame->data[3] && frame->linesize[3] )
1451 {
1452 int i;
1453 uint8_t *src, *dst;
1454
1455 dst = *alpha = mlt_pool_alloc( width * height );
1456 src = frame->data[3];
1457
1458 for ( i = 0; i < height; dst += width, src += frame->linesize[3], i++ )
1459 memcpy( dst, src, FFMIN( width, frame->linesize[3] ) );
1460 }
1461
1462 int src_pix_fmt = pix_fmt;
1463 pick_av_pixel_format( &src_pix_fmt );
1464 if ( *format == mlt_image_yuv420p )
1465 {
1466 // This is a special case. Movit wants the full range, if available.
1467 // Thankfully, there is not much other use of yuv420p except consumer
1468 // avformat with no filters and explicitly requested.
1469 #if defined(FFUDIV)
1470 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_YUV420P);
1471 struct SwsContext *context = sws_getContext(width, height, src_pix_fmt,
1472 width, height, AV_PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1473 #else
1474 int dst_pix_fmt = self->full_luma ? AV_PIX_FMT_YUVJ420P : AV_PIX_FMT_YUV420P;
1475 int flags = mlt_get_sws_flags(width, height, pix_fmt, width, height, dst_pix_fmt);
1476 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1477 width, height, dst_pix_fmt,
1478 flags, NULL, NULL, NULL);
1479 #endif
1480
1481 uint8_t *out_data[4];
1482 int out_stride[4];
1483 out_data[0] = buffer;
1484 out_data[1] = buffer + width * height;
1485 out_data[2] = buffer + ( 5 * width * height ) / 4;
1486 out_stride[0] = width;
1487 out_stride[1] = width >> 1;
1488 out_stride[2] = width >> 1;
1489 if ( !mlt_set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma, self->full_luma ) )
1490 result = profile->colorspace;
1491 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1492 out_data, out_stride);
1493 sws_freeContext( context );
1494 }
1495 else if ( *format == mlt_image_rgb )
1496 {
1497 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_RGB24);
1498 struct SwsContext *context = sws_getContext( width, height, src_pix_fmt,
1499 width, height, AV_PIX_FMT_RGB24, flags, NULL, NULL, NULL);
1500 uint8_t *out_data[4];
1501 int out_stride[4];
1502 av_image_fill_arrays(out_data, out_stride, buffer, AV_PIX_FMT_RGB24, width, height, IMAGE_ALIGN);
1503 // libswscale wants the RGB colorspace to be SWS_CS_DEFAULT, which is = SWS_CS_ITU601.
1504 mlt_set_luma_transfer( context, self->yuv_colorspace, 601, self->full_luma, 0 );
1505 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1506 out_data, out_stride);
1507 sws_freeContext( context );
1508 }
1509 else if ( *format == mlt_image_rgba )
1510 {
1511 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_RGBA);
1512 struct SwsContext *context = sws_getContext( width, height, src_pix_fmt,
1513 width, height, AV_PIX_FMT_RGBA, flags, NULL, NULL, NULL);
1514 uint8_t *out_data[4];
1515 int out_stride[4];
1516 av_image_fill_arrays(out_data, out_stride, buffer, AV_PIX_FMT_RGBA, width, height, IMAGE_ALIGN);
1517 // libswscale wants the RGB colorspace to be SWS_CS_DEFAULT, which is = SWS_CS_ITU601.
1518 mlt_set_luma_transfer( context, self->yuv_colorspace, 601, self->full_luma, 0 );
1519 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1520 out_data, out_stride);
1521 sws_freeContext( context );
1522 }
1523 else
1524 #if defined(FFUDIV) && (LIBSWSCALE_VERSION_INT >= ((3<<16)+(1<<8)+101))
1525 {
1526 int i, c;
1527 struct sliced_pix_fmt_conv_t ctx =
1528 {
1529 .width = width,
1530 .height = height,
1531 .frame = frame,
1532 .dst_format = AV_PIX_FMT_YUYV422,
1533 .src_colorspace = self->yuv_colorspace,
1534 .dst_colorspace = profile->colorspace,
1535 .src_full_range = self->full_luma,
1536 .dst_full_range = 0,
1537 };
1538 ctx.src_format = (self->full_luma && src_pix_fmt == AV_PIX_FMT_YUV422P) ? AV_PIX_FMT_YUVJ422P : src_pix_fmt;
1539 ctx.src_desc = av_pix_fmt_desc_get( ctx.src_format );
1540 ctx.dst_desc = av_pix_fmt_desc_get( ctx.dst_format );
1541 ctx.flags = mlt_get_sws_flags(width, height, ctx.src_format, width, height, ctx.dst_format);
1542
1543 av_image_fill_arrays(ctx.out_data, ctx.out_stride, buffer, ctx.dst_format, width, height, IMAGE_ALIGN);
1544
1545 int sliced = !getenv("MLT_AVFORMAT_SLICED_PIXFMT_DISABLE");
1546 if ( sliced ) {
1547 ctx.slice_w = ( width < 1000 )
1548 ? ( 256 >> frame->interlaced_frame )
1549 : ( 512 >> frame->interlaced_frame );
1550 } else {
1551 ctx.slice_w = width;
1552 }
1553
1554 c = ( width + ctx.slice_w - 1 ) / ctx.slice_w;
1555 int last_slice_w = width - ctx.slice_w * (c - 1);
1556
1557 if ( sliced && (last_slice_w % 8) == 0 && !(ctx.src_format == AV_PIX_FMT_YUV422P && last_slice_w % 16) ) {
1558 c *= frame->interlaced_frame ? 2 : 1;
1559 mlt_slices_run_normal( c, sliced_h_pix_fmt_conv_proc, &ctx );
1560 } else {
1561 c = frame->interlaced_frame ? 2 : 1;
1562 ctx.slice_w = width;
1563 for ( i = 0 ; i < c; i++ )
1564 sliced_h_pix_fmt_conv_proc( i, i, c, &ctx );
1565 }
1566
1567 result = profile->colorspace;
1568 }
1569 #else
1570 {
1571 #if defined(FFUDIV)
1572 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_YUYV422);
1573 struct SwsContext *context = sws_getContext( width, height, src_pix_fmt,
1574 width, height, AV_PIX_FMT_YUYV422, flags, NULL, NULL, NULL);
1575 #else
1576 int flags = mlt_get_sws_flags(width, height, pix_fmt, width, height, AV_PIX_FMT_YUYV422);
1577 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1578 width, height, AV_PIX_FMT_YUYV422, flags, NULL, NULL, NULL);
1579 #endif
1580 AVPicture output;
1581 avpicture_fill( &output, buffer, AV_PIX_FMT_YUYV422, width, height );
1582 if ( !mlt_set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma, 0 ) )
1583 result = profile->colorspace;
1584 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1585 output.data, output.linesize);
1586 sws_freeContext( context );
1587 }
1588 #endif
1589 mlt_log_timings_end( NULL, __FUNCTION__ );
1590
1591 return result;
1592 }
1593
set_image_size(producer_avformat self,int * width,int * height)1594 static void set_image_size( producer_avformat self, int *width, int *height )
1595 {
1596 double dar = mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE(self->parent) ) );
1597 double theta = self->autorotate? get_rotation( self->video_format->streams[self->video_index] ) : 0.0;
1598 if ( fabs(theta - 90.0) < 1.0 || fabs(theta - 270.0) < 1.0 )
1599 {
1600 *height = self->video_codec->width;
1601 // Workaround 1088 encodings missing cropping info.
1602 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 )
1603 *width = 1080;
1604 else
1605 *width = self->video_codec->height;
1606 } else {
1607 *width = self->video_codec->width;
1608 // Workaround 1088 encodings missing cropping info.
1609 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 )
1610 *height = 1080;
1611 else
1612 *height = self->video_codec->height;
1613 }
1614 }
1615
1616 /** Allocate the image buffer and set it on the frame.
1617 */
1618
allocate_buffer(mlt_frame frame,AVCodecParameters * codec_params,uint8_t ** buffer,mlt_image_format format,int width,int height)1619 static int allocate_buffer( mlt_frame frame, AVCodecParameters *codec_params, uint8_t **buffer, mlt_image_format format, int width, int height )
1620 {
1621 int size = 0;
1622
1623 if ( codec_params->width == 0 || codec_params->height == 0 )
1624 return size;
1625
1626 size = mlt_image_format_size( format, width, height, NULL );
1627 *buffer = mlt_pool_alloc( size );
1628 if ( *buffer )
1629 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1630 else
1631 size = 0;
1632
1633 return size;
1634 }
1635
ignore_send_packet_result(int result)1636 static int ignore_send_packet_result(int result)
1637 {
1638 return result >= 0 || result == AVERROR(EAGAIN) || result == AVERROR_EOF || result == AVERROR_INVALIDDATA;
1639 }
1640
1641 /** Get an image from a frame.
1642 */
1643
producer_get_image(mlt_frame frame,uint8_t ** buffer,mlt_image_format * format,int * width,int * height,int writable)1644 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1645 {
1646 // Get the producer
1647 (void) writable; // unused
1648 producer_avformat self = mlt_frame_pop_service( frame );
1649 mlt_producer producer = self->parent;
1650
1651 // Get the properties from the frame
1652 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1653
1654 // Obtain the frame number of this frame
1655 mlt_position position = mlt_frame_original_position( frame );
1656
1657 // Get the producer properties
1658 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1659
1660 uint8_t *alpha = NULL;
1661 int got_picture = 0;
1662 int image_size = 0;
1663
1664 pthread_mutex_lock( &self->video_mutex );
1665 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1666 mlt_log_timings_begin();
1667
1668 // Fetch the video format context
1669 AVFormatContext *context = self->video_format;
1670 AVCodecParameters *codec_params = NULL;
1671 if ( !context )
1672 goto exit_get_image;
1673
1674 // Get the video stream
1675 AVStream *stream = context->streams[ self->video_index ];
1676
1677 // Get codec context
1678 AVCodecContext *codec_context = stream->codec;
1679 codec_params = stream->codecpar;
1680
1681 // Always use the image cache for album art.
1682 int is_album_art = stream->disposition & AV_DISPOSITION_ATTACHED_PIC;
1683 if (is_album_art)
1684 position = 0;
1685
1686 // Get the image cache
1687 if ( ! self->image_cache )
1688 {
1689 // if cache size supplied by environment variable
1690 int cache_supplied = getenv( "MLT_AVFORMAT_CACHE" ) != NULL;
1691 int cache_size = cache_supplied? atoi( getenv( "MLT_AVFORMAT_CACHE" ) ) : 0;
1692
1693 // cache size supplied via property
1694 if ( mlt_properties_get( properties, "cache" ) )
1695 {
1696 cache_supplied = 1;
1697 cache_size = mlt_properties_get_int( properties, "cache" );
1698 }
1699 if ( mlt_properties_get_int( properties, "noimagecache" ) )
1700 {
1701 cache_supplied = 1;
1702 cache_size = 0;
1703 }
1704 // create cache if not disabled
1705 if ( !cache_supplied || cache_size > 0 )
1706 self->image_cache = mlt_cache_init();
1707 // set cache size if supplied
1708 if ( self->image_cache && cache_supplied )
1709 mlt_cache_set_size( self->image_cache, cache_size );
1710 }
1711 if ( self->image_cache )
1712 {
1713 mlt_frame original = mlt_cache_get_frame( self->image_cache, position );
1714 if ( original )
1715 {
1716 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
1717 int size = 0;
1718
1719 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
1720 if (*buffer)
1721 mlt_frame_set_alpha( frame, *buffer, size, NULL );
1722 *buffer = mlt_properties_get_data( orig_props, "image", &size );
1723 mlt_frame_set_image( frame, *buffer, size, NULL );
1724 mlt_properties_set_data( frame_properties, "avformat.image_cache", original, 0, (mlt_destructor) mlt_frame_close, NULL );
1725 *format = mlt_properties_get_int( orig_props, "format" );
1726 set_image_size( self, width, height );
1727 mlt_properties_pass_property(frame_properties, orig_props, "colorspace");
1728 got_picture = 1;
1729 goto exit_get_image;
1730 }
1731 }
1732 // Cache miss
1733
1734 // We may want to use the source fps if available
1735 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1736 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1737
1738 // This is the physical frame position in the source
1739 int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1740
1741 // Determines if we have to decode all frames in a sequence - when there temporal compression is used.
1742 const AVCodecDescriptor *descriptor = avcodec_descriptor_get( codec_params->codec_id );
1743 int must_decode = descriptor && !( descriptor->props & AV_CODEC_PROP_INTRA_ONLY );
1744
1745 double delay = mlt_properties_get_double( properties, "video_delay" );
1746
1747 // Seek if necessary
1748 double speed = mlt_producer_get_speed(producer);
1749 int preseek = must_decode && codec_context->has_b_frames && speed >= 0.0 && speed <= 1.0;
1750 int paused = seek_video( self, position, req_position, preseek );
1751
1752 // Seek might have reopened the file
1753 context = self->video_format;
1754 stream = context->streams[ self->video_index ];
1755 codec_context = stream->codec;
1756 codec_params = stream->codecpar;
1757 if ( *format == mlt_image_none || *format == mlt_image_movit ||
1758 codec_params->format == AV_PIX_FMT_ARGB ||
1759 codec_params->format == AV_PIX_FMT_RGBA ||
1760 codec_params->format == AV_PIX_FMT_ABGR ||
1761 codec_params->format == AV_PIX_FMT_BGRA )
1762 *format = pick_image_format( codec_params->format );
1763 #if defined(FFUDIV)
1764 else if ( codec_params->format == AV_PIX_FMT_BAYER_RGGB16LE ) {
1765 if ( *format == mlt_image_yuv422 )
1766 *format = mlt_image_yuv420p;
1767 else if ( *format == mlt_image_rgba )
1768 *format = mlt_image_rgb;
1769 }
1770 #endif
1771 else if ( codec_params->format == AV_PIX_FMT_YUVA444P10LE
1772 #if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(56,0,0)
1773 || codec_params->format == AV_PIX_FMT_GBRAP10LE
1774 || codec_params->format == AV_PIX_FMT_GBRAP12LE
1775 #endif
1776 )
1777 *format = mlt_image_rgba;
1778
1779 // Duplicate the last image if necessary
1780 if ( self->video_frame && self->video_frame->linesize[0]
1781 && (self->pkt.stream_index == self->video_index )
1782 && ( paused || self->current_position >= req_position ) )
1783 {
1784 // Duplicate it
1785 set_image_size( self, width, height );
1786 if ( ( image_size = allocate_buffer( frame, codec_params, buffer, *format, *width, *height ) ) )
1787 {
1788 int yuv_colorspace;
1789 #if USE_HWACCEL
1790 yuv_colorspace = convert_image( self, self->video_frame, *buffer, self->video_frame->format,
1791 format, *width, *height, &alpha );
1792 #else
1793 yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_params->format,
1794 format, *width, *height, &alpha );
1795 #endif
1796 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1797 got_picture = 1;
1798 }
1799 }
1800 else
1801 {
1802 int64_t int_position = 0;
1803 int decode_errors = 0;
1804
1805 // Construct an AVFrame for YUV422 conversion
1806 if ( !self->video_frame )
1807 self->video_frame = av_frame_alloc();
1808 else
1809 av_frame_unref( self->video_frame );
1810
1811 while (!got_picture && ignore_send_packet_result(self->video_send_result))
1812 {
1813 if ( self->video_send_result != AVERROR( EAGAIN ) )
1814 {
1815 // Read a packet
1816 if ( self->pkt.stream_index == self->video_index )
1817 av_packet_unref( &self->pkt );
1818 av_init_packet( &self->pkt );
1819 pthread_mutex_lock( &self->packets_mutex );
1820 if ( mlt_deque_count( self->vpackets ) )
1821 {
1822 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1823 av_packet_ref( &self->pkt, tmp );
1824 av_packet_free( &tmp );
1825 }
1826 else
1827 {
1828 int ret = av_read_frame( context, &self->pkt );
1829 if ( ret >= 0 && !self->video_seekable && self->pkt.stream_index == self->audio_index )
1830 {
1831 mlt_deque_push_back( self->apackets, av_packet_clone( &self->pkt ) );
1832 }
1833 else if ( ret < 0 )
1834 {
1835 if ( ret == AVERROR_EOF )
1836 {
1837 self->pkt.stream_index = self->video_index;
1838 } else
1839 {
1840 mlt_log_verbose( MLT_PRODUCER_SERVICE( producer ), "av_read_frame returned error %d inside get_image\n", ret );
1841 }
1842 if ( !self->video_seekable && mlt_properties_get_int( properties, "reconnect" ) )
1843 {
1844 // Try to reconnect to live sources by closing context and codecs,
1845 // and letting next call to get_frame() reopen.
1846 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1847 prepare_reopen( self );
1848 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1849 pthread_mutex_unlock( &self->packets_mutex );
1850 goto exit_get_image;
1851 }
1852 if ( !self->video_seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) )
1853 {
1854 mlt_log_fatal( MLT_PRODUCER_SERVICE( producer ), "Exiting with error due to disconnected source.\n" );
1855 exit( EXIT_FAILURE );
1856 }
1857 // Send null packets to drain decoder.
1858 self->pkt.size = 0;
1859 self->pkt.data = NULL;
1860 }
1861 }
1862 pthread_mutex_unlock( &self->packets_mutex );
1863 }
1864
1865 // We only deal with video from the selected video_index
1866 if ( self->pkt.stream_index == self->video_index )
1867 {
1868 int64_t pts = best_pts( self, self->pkt.pts, self->pkt.dts );
1869 if ( pts != AV_NOPTS_VALUE )
1870 {
1871 if ( !self->video_seekable && self->first_pts == AV_NOPTS_VALUE )
1872 self->first_pts = pts;
1873 if ( self->first_pts != AV_NOPTS_VALUE )
1874 pts -= self->first_pts;
1875 else if ( context->start_time != AV_NOPTS_VALUE )
1876 pts -= context->start_time;
1877 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1878 if ( int_position == self->last_position )
1879 int_position = self->last_position + 1;
1880 }
1881 mlt_log_debug( MLT_PRODUCER_SERVICE(producer),
1882 "V pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1883 self->pkt.pts, self->pkt.dts, req_position, self->current_position, int_position );
1884
1885 // Make a dumb assumption on streams that contain wild timestamps
1886 if ( llabs( req_position - int_position ) > 999 )
1887 {
1888 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP: "
1889 "pkt.pts=[%"PRId64"], pkt.dts=[%"PRId64"], req_position=[%"PRId64"], "
1890 "current_position=[%"PRId64"], int_position=[%"PRId64"], pts=[%"PRId64"] \n",
1891 self->pkt.pts, self->pkt.dts, req_position,
1892 self->current_position, int_position, pts );
1893 int_position = req_position;
1894 }
1895 self->last_position = int_position;
1896
1897 // Decode the image
1898 if ( must_decode || int_position >= req_position || !self->pkt.data )
1899 {
1900 codec_context->reordered_opaque = int_position;
1901 if ( int_position >= req_position )
1902 codec_context->skip_loop_filter = AVDISCARD_NONE;
1903 self->video_send_result = avcodec_send_packet( codec_context, &self->pkt );
1904 mlt_log_debug( MLT_PRODUCER_SERVICE( producer ), "decoded video packet with size %d => %d\n", self->pkt.size, self->video_send_result );
1905 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1906 if (!ignore_send_packet_result(self->video_send_result))
1907 {
1908 mlt_log_warning( MLT_PRODUCER_SERVICE( producer ), "video avcodec_send_packet failed with %d\n", self->video_send_result );
1909 }
1910 else
1911 {
1912 int error = avcodec_receive_frame( codec_context, self->video_frame );
1913 if ( error < 0 )
1914 {
1915 if ( error != AVERROR( EAGAIN ) && ++decode_errors > 10 )
1916 {
1917 mlt_log_warning( MLT_PRODUCER_SERVICE( producer ), "video decoding error %d\n", error );
1918 self->last_good_position = POSITION_INVALID;
1919 }
1920 }
1921 else
1922 {
1923 #if USE_HWACCEL
1924 if (self->hwaccel.device_ctx && self->video_frame->format == self->hwaccel.pix_fmt)
1925 {
1926 AVFrame *sw_video_frame = av_frame_alloc();
1927 int transfer_data_result = av_hwframe_transfer_data(sw_video_frame, self->video_frame, 0);
1928 if(transfer_data_result < 0)
1929 {
1930 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "av_hwframe_transfer_data() failed %d\n", transfer_data_result);
1931 av_frame_free(&sw_video_frame);
1932 goto exit_get_image;
1933 }
1934 av_frame_copy_props(sw_video_frame, self->video_frame);
1935 sw_video_frame->width = self->video_frame->width;
1936 sw_video_frame->height = self->video_frame->height;
1937
1938 av_frame_unref(self->video_frame);
1939 av_frame_move_ref(self->video_frame, sw_video_frame);
1940 av_frame_free(&sw_video_frame);
1941 }
1942 #endif
1943 got_picture = 1;
1944 decode_errors = 0;
1945 }
1946 }
1947 }
1948
1949 if ( got_picture )
1950 {
1951 // Get position of reordered frame
1952 int_position = self->video_frame->reordered_opaque;
1953 pts = best_pts( self, self->video_frame->pts, self->video_frame->pkt_dts );
1954 if ( pts != AV_NOPTS_VALUE )
1955 {
1956 // Some streams are not marking their key frames even though
1957 // there are I frames, and find_first_pts() fails as a result.
1958 // Try to set first_pts here after getting pict_type.
1959 if ( self->first_pts == AV_NOPTS_VALUE &&
1960 (self->video_frame->key_frame || self->video_frame->pict_type == AV_PICTURE_TYPE_I) )
1961 self->first_pts = pts;
1962 if ( self->first_pts != AV_NOPTS_VALUE )
1963 pts -= self->first_pts;
1964 else if ( context->start_time != AV_NOPTS_VALUE )
1965 pts -= context->start_time;
1966 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1967 }
1968
1969 if ( int_position < req_position )
1970 got_picture = 0;
1971 else if ( int_position >= req_position )
1972 codec_context->skip_loop_filter = AVDISCARD_NONE;
1973 }
1974 else if ( !self->pkt.data ) // draining decoder with null packets
1975 {
1976 self->video_send_result = -1;
1977 }
1978 mlt_log_debug( MLT_PRODUCER_SERVICE( producer ), " got_pic %d key %d send_result %d pkt_pos %"PRId64"\n",
1979 got_picture, self->pkt.flags & AV_PKT_FLAG_KEY, self->video_send_result, int_position );
1980 }
1981
1982 // Now handle the picture if we have one
1983 if ( got_picture )
1984 {
1985 #ifdef AVFILTER
1986 if (self->autorotate && self->vfilter_graph) {
1987 int ret = av_buffersrc_add_frame(self->vfilter_in, self->video_frame);
1988 if (ret < 0) {
1989 got_picture = 0;
1990 break;
1991 }
1992 while (ret >= 0) {
1993 ret = av_buffersink_get_frame_flags(self->vfilter_out, self->video_frame, 0);
1994 if (ret < 0) {
1995 ret = 0;
1996 break;
1997 }
1998 }
1999 }
2000 #endif
2001 set_image_size( self, width, height );
2002 if ( ( image_size = allocate_buffer( frame, codec_params, buffer, *format, *width, *height ) ) )
2003 {
2004 int yuv_colorspace;
2005 #if USE_HWACCEL
2006 // not sure why this is really needed, but doesn't seem to work otherwise
2007 yuv_colorspace = convert_image( self, self->video_frame, *buffer, self->video_frame->format,
2008 format, *width, *height, &alpha );
2009 #else
2010 yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_params->format,
2011 format, *width, *height, &alpha );
2012 #endif
2013 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
2014 self->top_field_first |= self->video_frame->top_field_first;
2015 self->top_field_first |= codec_params->field_order == AV_FIELD_TT;
2016 self->top_field_first |= codec_params->field_order == AV_FIELD_TB;
2017 self->current_position = int_position;
2018 }
2019 else
2020 {
2021 got_picture = 0;
2022 }
2023 }
2024
2025 // Free packet data if not video and not live audio packet
2026 if ( self->pkt.stream_index != self->video_index &&
2027 !( !self->video_seekable && self->pkt.stream_index == self->audio_index ) )
2028 av_packet_unref( &self->pkt );
2029 }
2030 }
2031
2032 // set alpha
2033 if ( alpha )
2034 mlt_frame_set_alpha( frame, alpha, (*width) * (*height), mlt_pool_release );
2035
2036 if ( image_size > 0 )
2037 {
2038 mlt_properties_set_int( frame_properties, "format", *format );
2039 // Cache the image for rapid repeated access.
2040 if ( self->image_cache ) {
2041 if (is_album_art) {
2042 mlt_position original_pos = mlt_frame_original_position( frame );
2043 mlt_properties_set_position(frame_properties, "original_position", 0);
2044 mlt_cache_put_frame( self->image_cache, frame );
2045 mlt_properties_set_position(frame_properties, "original_position", original_pos);
2046 } else {
2047 mlt_cache_put_frame( self->image_cache, frame );
2048 }
2049 }
2050 // Clone frame for error concealment.
2051 if ( self->current_position >= self->last_good_position ) {
2052 self->last_good_position = self->current_position;
2053 if ( self->last_good_frame )
2054 mlt_frame_close( self->last_good_frame );
2055 self->last_good_frame = mlt_frame_clone( frame, 1 );
2056 }
2057 }
2058 else if ( self->last_good_frame )
2059 {
2060 // Use last known good frame if there was a decoding failure.
2061 mlt_frame original = mlt_frame_clone( self->last_good_frame, 1 );
2062 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
2063 int size = 0;
2064
2065 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
2066 if (*buffer)
2067 mlt_frame_set_alpha( frame, *buffer, size, NULL );
2068 *buffer = mlt_properties_get_data( orig_props, "image", &size );
2069 mlt_frame_set_image( frame, *buffer, size, NULL );
2070 mlt_properties_set_data( frame_properties, "avformat.conceal_error", original, 0, (mlt_destructor) mlt_frame_close, NULL );
2071 *format = mlt_properties_get_int( orig_props, "format" );
2072 set_image_size( self, width, height );
2073 got_picture = 1;
2074 }
2075
2076 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
2077 self->video_expected = position + 1;
2078
2079 exit_get_image:
2080 pthread_mutex_unlock( &self->video_mutex );
2081
2082 // Set the progressive flag
2083 if ( mlt_properties_get( properties, "force_progressive" ) ) {
2084 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
2085 } else if ( self->video_frame && codec_params ) {
2086 mlt_properties_set_int( frame_properties, "progressive",
2087 !self->video_frame->interlaced_frame &&
2088 (codec_params->field_order == AV_FIELD_PROGRESSIVE ||
2089 codec_params->field_order == AV_FIELD_UNKNOWN) );
2090 }
2091
2092 // Set the field order property for this frame
2093 if ( mlt_properties_get( properties, "force_tff" ) )
2094 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
2095 else
2096 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
2097
2098 // Set immutable properties of the selected track's (or overridden) source attributes.
2099 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
2100 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
2101 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
2102
2103 mlt_log_timings_end( NULL, __FUNCTION__ );
2104
2105 return !got_picture;
2106 }
2107
2108 /** Process properties as AVOptions and apply to AV context obj
2109 */
2110
apply_properties(void * obj,mlt_properties properties,int flags)2111 static void apply_properties( void *obj, mlt_properties properties, int flags )
2112 {
2113 int i;
2114 int count = mlt_properties_count( properties );
2115 for ( i = 0; i < count; i++ )
2116 {
2117 const char *opt_name = mlt_properties_get_name( properties, i );
2118 int search_flags = AV_OPT_SEARCH_CHILDREN;
2119 const AVOption *opt = av_opt_find( obj, opt_name, NULL, flags, search_flags );
2120 if ( opt_name && mlt_properties_get( properties, opt_name ) && strcmp(opt_name, "seekable") )
2121 {
2122 if ( opt )
2123 av_opt_set( obj, opt_name, mlt_properties_get( properties, opt_name), search_flags );
2124 }
2125 }
2126 }
2127
2128 /** Initialize the video codec context.
2129 */
2130
video_codec_init(producer_avformat self,int index,mlt_properties properties)2131 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
2132 {
2133 // Initialise the codec if necessary
2134 if ( !self->video_codec )
2135 {
2136 // Get the video stream
2137 AVStream *stream = self->video_format->streams[ index ];
2138
2139 // Get codec context
2140 AVCodecContext *codec_context = stream->codec;
2141 AVCodecParameters *codec_params = stream->codecpar;
2142
2143 // Find the codec
2144 AVCodec *codec = avcodec_find_decoder( codec_params->codec_id );
2145 if ( mlt_properties_get( properties, "vcodec" ) ) {
2146 if ( !( codec = avcodec_find_decoder_by_name( mlt_properties_get( properties, "vcodec" ) ) ) )
2147 codec = avcodec_find_decoder( codec_params->codec_id );
2148 } else if ( codec_params->codec_id == AV_CODEC_ID_VP9 ) {
2149 if ( !( codec = avcodec_find_decoder_by_name( "libvpx-vp9" ) ) )
2150 codec = avcodec_find_decoder( codec_params->codec_id );
2151 } else if ( codec_params->codec_id == AV_CODEC_ID_VP8 ) {
2152 if ( !( codec = avcodec_find_decoder_by_name( "libvpx" ) ) )
2153 codec = avcodec_find_decoder( codec_params->codec_id );
2154 }
2155
2156 // Initialise multi-threading
2157 int thread_count = mlt_properties_get_int( properties, "threads" );
2158 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
2159 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
2160 if ( thread_count >= 0 )
2161 codec_context->thread_count = thread_count;
2162
2163 #if USE_HWACCEL
2164 if ( self->hwaccel.device_type == AV_HWDEVICE_TYPE_NONE || self->hwaccel.pix_fmt == AV_PIX_FMT_NONE )
2165 {
2166 mlt_log_verbose( MLT_PRODUCER_SERVICE( self->parent ), "missing hwaccel parameters. skipping hardware initialization\n" );
2167 goto skip_hwaccel;
2168 }
2169
2170 int found_hw_pix_fmt = 0, i;
2171 for ( i = 0;; i++ )
2172 {
2173 const AVCodecHWConfig *config = avcodec_get_hw_config( codec, i );
2174 if ( !config )
2175 break;
2176
2177 if ( config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX &&
2178 config->device_type == self->hwaccel.device_type && config->pix_fmt == self->hwaccel.pix_fmt )
2179 {
2180 found_hw_pix_fmt = 1;
2181 break;
2182 }
2183 }
2184
2185 if ( found_hw_pix_fmt )
2186 {
2187 av_buffer_unref( &self->hwaccel.device_ctx );
2188 int ret = av_hwdevice_ctx_create( &self->hwaccel.device_ctx, self->hwaccel.device_type, self->hwaccel.device, NULL, 0 );
2189 if ( ret >= 0 )
2190 {
2191 codec_context->hw_device_ctx = av_buffer_ref( self->hwaccel.device_ctx );
2192 mlt_log_info( MLT_PRODUCER_SERVICE( self->parent ), "av_hwdevice_ctx_create() success %d\n", codec_context->pix_fmt );
2193 }
2194 else
2195 {
2196 mlt_log_warning( MLT_PRODUCER_SERVICE( self->parent ), "av_hwdevice_ctx_create() failed %d\n", ret );
2197 }
2198 }
2199 else
2200 {
2201 mlt_log_warning( MLT_PRODUCER_SERVICE( self->parent ), "failed to find hw_pix_fmt\n" );
2202 }
2203
2204 skip_hwaccel:
2205 #endif
2206 // If we don't have a codec and we can't initialise it, we can't do much more...
2207 pthread_mutex_lock( &self->open_mutex );
2208 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2209 {
2210 // Switch to the native vp8/vp9 decoder if not yuva420p
2211 if ( codec_params->format != AV_PIX_FMT_YUVA420P
2212 && !mlt_properties_get( properties, "vcodec" )
2213 && ( !strcmp(codec->name, "libvpx") || !strcmp(codec->name, "libvpx-vp9") ) )
2214 {
2215 codec = avcodec_find_decoder( codec_params->codec_id );
2216 if ( codec && avcodec_open2( codec_context, codec, NULL ) < 0 ) {
2217 self->video_index = -1;
2218 pthread_mutex_unlock( &self->open_mutex );
2219 return 0;
2220 }
2221 }
2222 // Now store the codec with its destructor
2223 self->video_codec = codec_context;
2224 }
2225 else
2226 {
2227 // Remember that we can't use this later
2228 self->video_index = -1;
2229 pthread_mutex_unlock( &self->open_mutex );
2230 return 0;
2231 }
2232 pthread_mutex_unlock( &self->open_mutex );
2233
2234 // Process properties as AVOptions
2235 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2236 if ( codec && codec->priv_class && codec_context->priv_data )
2237 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2238
2239 // Reset some image properties
2240 mlt_properties_set_int( properties, "width", codec_params->width );
2241 mlt_properties_set_int( properties, "height", codec_params->height );
2242 get_aspect_ratio( properties, stream, codec_params );
2243
2244 // Start with the muxer frame rate.
2245 AVRational frame_rate = stream->avg_frame_rate;
2246 double fps = av_q2d( frame_rate );
2247
2248 #if defined(FFUDIV)
2249 // Verify and sanitize the muxer frame rate.
2250 if ( isnan( fps ) || isinf( fps ) || fps == 0 )
2251 {
2252 frame_rate = stream->r_frame_rate;
2253 fps = av_q2d( frame_rate );
2254 }
2255 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
2256 // avg_frame_rate gives some approximate value that does not well match the media.
2257 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
2258 // results in some very choppy output, but some value slightly different works
2259 // great.
2260 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
2261 {
2262 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
2263 fps = av_q2d( frame_rate );
2264 }
2265 #endif
2266 // XXX frame rates less than 1 fps are not considered sane
2267 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
2268 {
2269 // Get the frame rate from the codec.
2270 frame_rate.num = self->video_codec->time_base.den;
2271 frame_rate.den = self->video_codec->time_base.num * self->video_codec->ticks_per_frame;
2272 fps = av_q2d( frame_rate );
2273 }
2274 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
2275 {
2276 // Use the profile frame rate if all else fails.
2277 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
2278 frame_rate.num = profile->frame_rate_num;
2279 frame_rate.den = profile->frame_rate_den;
2280 }
2281
2282 self->video_time_base = stream->time_base;
2283 if ( mlt_properties_get( properties, "force_fps" ) )
2284 {
2285 AVRational force_fps = av_d2q( mlt_properties_get_double( properties, "force_fps" ), 1024 );
2286 self->video_time_base = av_mul_q( stream->time_base, av_div_q( frame_rate, force_fps ) );
2287 frame_rate = force_fps;
2288 }
2289 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
2290 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
2291
2292 // Cover art is a single image at 90000 fps, which is not seekable.
2293 if (stream->disposition & AV_DISPOSITION_ATTACHED_PIC)
2294 self->video_seekable = 0;
2295
2296 // Set the YUV colorspace from override or detect
2297 self->yuv_colorspace = mlt_properties_get_int( properties, "force_colorspace" );
2298 if ( ! self->yuv_colorspace )
2299 {
2300 switch ( self->video_codec->colorspace )
2301 {
2302 case AVCOL_SPC_SMPTE240M:
2303 self->yuv_colorspace = 240;
2304 break;
2305 case AVCOL_SPC_BT470BG:
2306 case AVCOL_SPC_SMPTE170M:
2307 self->yuv_colorspace = 601;
2308 break;
2309 case AVCOL_SPC_BT709:
2310 self->yuv_colorspace = 709;
2311 break;
2312 default:
2313 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
2314 self->yuv_colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
2315 break;
2316 }
2317 }
2318 // Let apps get chosen colorspace
2319 mlt_properties_set_int( properties, "meta.media.colorspace", self->yuv_colorspace );
2320
2321 // Get the color transfer characteristic (gamma).
2322 self->color_trc = mlt_properties_get_int( properties, "force_color_trc" );
2323 if ( !self->color_trc )
2324 self->color_trc = self->video_codec->color_trc;
2325 mlt_properties_set_int( properties, "meta.media.color_trc", self->color_trc );
2326
2327 // Get the RGB color primaries.
2328 switch ( self->video_codec->color_primaries )
2329 {
2330 case AVCOL_PRI_BT470BG:
2331 self->color_primaries = 601625;
2332 break;
2333 case AVCOL_PRI_SMPTE170M:
2334 case AVCOL_PRI_SMPTE240M:
2335 self->color_primaries = 601525;
2336 break;
2337 case AVCOL_PRI_BT709:
2338 case AVCOL_PRI_UNSPECIFIED:
2339 default:
2340 self->color_primaries = 709;
2341 break;
2342 }
2343
2344 mlt_properties_set_int( properties, "meta.media.has_b_frames", self->video_codec->has_b_frames );
2345
2346 self->full_luma = 0;
2347 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "color_range %d\n", codec_context->color_range );
2348 if ( codec_context->color_range == AVCOL_RANGE_JPEG )
2349 self->full_luma = 1;
2350 if ( mlt_properties_get( properties, "set.force_full_luma" ) )
2351 self->full_luma = mlt_properties_get_int( properties, "set.force_full_luma" );
2352 }
2353 return self->video_index > -1;
2354 }
2355
2356 /** Set up video handling.
2357 */
2358
producer_set_up_video(producer_avformat self,mlt_frame frame)2359 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
2360 {
2361 // Get the producer
2362 mlt_producer producer = self->parent;
2363
2364 // Get the properties
2365 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2366
2367 // Fetch the video format context
2368 AVFormatContext *context = self->video_format;
2369
2370 // Get the video_index
2371 int index = mlt_properties_get_int( properties, "video_index" );
2372
2373 int unlock_needed = 0;
2374
2375 // Reopen the file if necessary
2376 if ( !context && index > -1 )
2377 {
2378 unlock_needed = 1;
2379 pthread_mutex_lock( &self->video_mutex );
2380 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2381 mlt_properties_get( properties, "resource" ), 0, 0 );
2382 context = self->video_format;
2383 }
2384
2385 // Exception handling for video_index
2386 if ( context && index >= (int) context->nb_streams )
2387 {
2388 // Get the last video stream
2389 for ( index = context->nb_streams - 1;
2390 index >= 0 && context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_VIDEO;
2391 index-- );
2392 mlt_properties_set_int( properties, "video_index", index );
2393 }
2394 if ( context && index > -1 && context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_VIDEO )
2395 {
2396 // Invalidate the video stream
2397 index = -1;
2398 mlt_properties_set_int( properties, "video_index", index );
2399 }
2400
2401 // Update the video properties if the index changed
2402 if ( context && index > -1 && index != self->video_index )
2403 {
2404 // Reset the video properties if the index changed
2405 self->video_index = index;
2406 pthread_mutex_lock( &self->open_mutex );
2407 if ( self->video_codec )
2408 avcodec_close( self->video_codec );
2409 self->video_codec = NULL;
2410 pthread_mutex_unlock( &self->open_mutex );
2411 }
2412
2413 // Get the frame properties
2414 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2415
2416 // Get the codec
2417 if ( context && index > -1 && video_codec_init( self, index, properties ) )
2418 {
2419 // Set the frame properties
2420 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
2421 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
2422 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
2423
2424 // Set the width and height
2425 double dar = mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) );
2426 double theta = self->autorotate? get_rotation( self->video_format->streams[index] ) : 0.0;
2427 if ( fabs(theta - 90.0) < 1.0 || fabs(theta - 270.0) < 1.0 )
2428 {
2429 // Workaround 1088 encodings missing cropping info.
2430 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 ) {
2431 mlt_properties_set_int( frame_properties, "width", 1080 );
2432 mlt_properties_set_int( properties, "meta.media.width", 1080 );
2433 } else {
2434 mlt_properties_set_int( frame_properties, "width", self->video_codec->height );
2435 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->height );
2436 }
2437 mlt_properties_set_int( frame_properties, "height", self->video_codec->width );
2438 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->width );
2439 aspect_ratio = ( force_aspect_ratio > 0.0 ) ? force_aspect_ratio : 1.0 / aspect_ratio;
2440 mlt_properties_set_double( frame_properties, "aspect_ratio", 1.0/aspect_ratio );
2441 } else {
2442 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
2443 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
2444 // Workaround 1088 encodings missing cropping info.
2445 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 ) {
2446 mlt_properties_set_int( frame_properties, "height", 1080 );
2447 mlt_properties_set_int( properties, "meta.media.height", 1080 );
2448 } else {
2449 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
2450 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
2451 }
2452 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
2453 }
2454 mlt_properties_set_int( frame_properties, "colorspace", self->yuv_colorspace );
2455 mlt_properties_set_int( frame_properties, "color_trc", self->color_trc );
2456 mlt_properties_set_int( frame_properties, "color_primaries", self->color_primaries );
2457 mlt_properties_set_int( frame_properties, "full_luma", self->full_luma );
2458 mlt_properties_set( properties, "meta.media.color_range", self->full_luma? "full" : "mpeg" );
2459
2460 // Add our image operation
2461 mlt_frame_push_service( frame, self );
2462 mlt_frame_push_get_image( frame, producer_get_image );
2463 }
2464 else
2465 {
2466 // If something failed, use test card image
2467 mlt_properties_set_int( frame_properties, "test_image", 1 );
2468 }
2469 if ( unlock_needed )
2470 pthread_mutex_unlock( &self->video_mutex );
2471 }
2472
seek_audio(producer_avformat self,mlt_position position,double timecode)2473 static int seek_audio( producer_avformat self, mlt_position position, double timecode )
2474 {
2475 int paused = 0;
2476
2477 pthread_mutex_lock( &self->packets_mutex );
2478
2479 // Seek if necessary
2480 if ( self->seekable && ( position != self->audio_expected || self->last_position < 0 ) )
2481 {
2482 if ( self->last_position == POSITION_INITIAL )
2483 {
2484 int video_index = self->video_index;
2485 if ( video_index == -1 )
2486 video_index = first_video_index( self );
2487 if ( self->first_pts == AV_NOPTS_VALUE && video_index >= 0 )
2488 find_first_pts( self, video_index );
2489 }
2490
2491 if ( position + 1 == self->audio_expected &&
2492 mlt_properties_get_int( MLT_PRODUCER_PROPERTIES( self->parent ), "mute_on_pause" ) )
2493 {
2494 // We're paused - silence required
2495 paused = 1;
2496 }
2497 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
2498 {
2499 AVFormatContext *context = self->audio_format;
2500 int64_t timestamp = llrint( timecode * AV_TIME_BASE );
2501 if ( context->start_time != AV_NOPTS_VALUE )
2502 timestamp += context->start_time;
2503 if ( timestamp < 0 )
2504 timestamp = 0;
2505
2506 // Set to the real timecode
2507 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
2508 paused = 1;
2509
2510 // Clear the usage in the audio buffer
2511 int i = MAX_AUDIO_STREAMS + 1;
2512 while ( --i )
2513 self->audio_used[i - 1] = 0;
2514 }
2515 }
2516 pthread_mutex_unlock( &self->packets_mutex );
2517 return paused;
2518 }
2519
sample_bytes(AVCodecContext * context)2520 static int sample_bytes( AVCodecContext *context )
2521 {
2522 return av_get_bytes_per_sample( context->sample_fmt );
2523 }
2524
planar_to_interleaved(uint8_t * dest,AVFrame * src,int samples,int channels,int bytes_per_sample)2525 static void planar_to_interleaved( uint8_t *dest, AVFrame *src, int samples, int channels, int bytes_per_sample )
2526 {
2527 int s, c;
2528 for ( s = 0; s < samples; s++ )
2529 {
2530 for ( c = 0; c < channels; c++ )
2531 {
2532 if ( c < AV_NUM_DATA_POINTERS )
2533 memcpy( dest, &src->data[c][s * bytes_per_sample], bytes_per_sample );
2534 dest += bytes_per_sample;
2535 }
2536 }
2537 }
2538
decode_audio(producer_avformat self,int * ignore,const AVPacket * pkt,int samples,double timecode,double fps)2539 static int decode_audio( producer_avformat self, int *ignore, const AVPacket *pkt, int samples, double timecode, double fps )
2540 {
2541 // Fetch the audio_format
2542 AVFormatContext *context = self->audio_format;
2543
2544 // Get the current stream index
2545 int index = pkt->stream_index;
2546
2547 // Get codec context
2548 AVCodecContext *codec_context = self->audio_codec[ index ];
2549
2550 // Obtain the audio buffers
2551 uint8_t *audio_buffer = self->audio_buffer[ index ];
2552
2553 int channels = codec_context->channels;
2554 int audio_used = self->audio_used[ index ];
2555 int ret = 0;
2556 int discarded = 1;
2557 int sizeof_sample = sample_bytes( codec_context );
2558
2559 // Decode the audio
2560 if ( !self->audio_frame )
2561 self->audio_frame = av_frame_alloc();
2562 else
2563 av_frame_unref( self->audio_frame );
2564 int error = avcodec_send_packet(codec_context, pkt);
2565 mlt_log_debug(MLT_PRODUCER_SERVICE(self->parent), "decoded audio packet with size %d => %d\n", pkt->size, error);
2566 if (!ignore_send_packet_result(error)) {
2567 mlt_log_warning(MLT_PRODUCER_SERVICE(self->parent), "audio avcodec_send_packet failed with %d\n", error);
2568 } else while (!error) {
2569 error = avcodec_receive_frame(codec_context, self->audio_frame);
2570 if (error) {
2571 if (error != AVERROR(EAGAIN)) {
2572 mlt_log_warning(MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", error);
2573 }
2574 } else {
2575 // Figure out how many samples will be needed after resampling
2576 int convert_samples = self->audio_frame->nb_samples;
2577 channels = codec_context->channels;
2578 ret += convert_samples * channels * sizeof_sample;
2579
2580 // Resize audio buffer to prevent overflow
2581 if ( ( audio_used + convert_samples ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2582 {
2583 self->audio_buffer_size[ index ] = ( audio_used + convert_samples * 2 ) * channels * sizeof_sample;
2584 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2585 }
2586 uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
2587 switch ( codec_context->sample_fmt )
2588 {
2589 case AV_SAMPLE_FMT_U8P:
2590 case AV_SAMPLE_FMT_S16P:
2591 case AV_SAMPLE_FMT_S32P:
2592 case AV_SAMPLE_FMT_FLTP:
2593 planar_to_interleaved( dest, self->audio_frame, convert_samples, channels, sizeof_sample );
2594 break;
2595 default: {
2596 int data_size = av_samples_get_buffer_size( NULL, channels,
2597 self->audio_frame->nb_samples, codec_context->sample_fmt, 1 );
2598 // Straight copy to audio buffer
2599 memcpy( dest, self->audio_frame->data[0], data_size );
2600 }
2601 }
2602 audio_used += convert_samples;
2603 discarded = 0;
2604 }
2605 }
2606
2607 // Handle ignore
2608 if ( *ignore > 0 && audio_used )
2609 {
2610 int n = FFMIN( audio_used, *ignore );
2611 *ignore -= n;
2612 audio_used -= n;
2613 memmove( audio_buffer, &audio_buffer[ n * channels * sizeof_sample ],
2614 audio_used * channels * sizeof_sample );
2615 }
2616
2617 // If we're behind, ignore this packet
2618 // Skip this on non-seekable, audio-only inputs.
2619 if ( !discarded && pkt->pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 )
2620 {
2621 int64_t pts = pkt->pts;
2622 if ( self->first_pts != AV_NOPTS_VALUE )
2623 pts -= self->first_pts;
2624 else if ( context->start_time != AV_NOPTS_VALUE && self->video_index != -1 )
2625 pts -= context->start_time;
2626 double timebase = av_q2d( context->streams[ index ]->time_base );
2627 int64_t int_position = llrint( timebase * pts * fps );
2628 int64_t req_position = llrint( timecode * fps );
2629 int64_t req_pts = llrint( timecode / timebase );
2630
2631 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
2632 "A pkt.pts %"PRId64" pkt->dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
2633 pkt->pts, pkt->dts, req_position, self->current_position, int_position );
2634
2635 if ( self->seekable || int_position > 0 )
2636 {
2637 if ( req_position > int_position ) {
2638 // We are behind, so skip some
2639 *ignore = lrint( timebase * (req_pts - pts) * codec_context->sample_rate );
2640 } else if ( self->audio_index != INT_MAX && int_position > req_position + 2 && !self->is_audio_synchronizing ) {
2641 // We are ahead, so seek backwards some more.
2642 // Supply -1 as the position to defeat the checks needed by for the other
2643 // call to seek_audio() at the beginning of producer_get_audio(). Otherwise,
2644 // more often than not, req_position will equal audio_expected.
2645 seek_audio( self, -1, timecode - 1.0 );
2646 self->is_audio_synchronizing = 1;
2647 }
2648 }
2649
2650 // Cancel the find_first_pts() in seek_audio()
2651 if ( self->video_index == -1 && self->last_position == POSITION_INITIAL )
2652 self->last_position = int_position;
2653 }
2654
2655 self->audio_used[ index ] = audio_used;
2656
2657 return ret;
2658 }
2659
2660 /** Get the audio from a frame.
2661 */
producer_get_audio(mlt_frame frame,void ** buffer,mlt_audio_format * format,int * frequency,int * channels,int * samples)2662 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2663 {
2664 // Get the producer
2665 producer_avformat self = mlt_frame_pop_audio( frame );
2666
2667 pthread_mutex_lock( &self->audio_mutex );
2668
2669 // Obtain the frame number of this frame
2670 mlt_position position = mlt_frame_original_position( frame );
2671
2672 // Calculate the real time code
2673 double real_timecode = producer_time_of_frame( self->parent, position );
2674
2675 // Get the producer fps
2676 double fps = mlt_producer_get_fps( self->parent );
2677 if ( mlt_properties_get( MLT_FRAME_PROPERTIES(frame), "producer_consumer_fps" ) )
2678 fps = mlt_properties_get_double( MLT_FRAME_PROPERTIES(frame), "producer_consumer_fps" );
2679
2680 // Number of frames to ignore (for ffwd)
2681 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2682
2683 // Flag for paused (silence)
2684 double timecode = self->audio_expected > 0 ? real_timecode : FFMAX(real_timecode - 0.25, 0.0);
2685 int paused = seek_audio( self, position, timecode );
2686
2687 // Initialize ignore for all streams from the seek return value
2688 int i = MAX_AUDIO_STREAMS;
2689 while ( i-- )
2690 ignore[i] = ignore[0];
2691
2692 // Fetch the audio_format
2693 AVFormatContext *context = self->audio_format;
2694 if ( !context )
2695 goto exit_get_audio;
2696
2697 int sizeof_sample = sizeof( int16_t );
2698
2699 // Determine the tracks to use
2700 int index = self->audio_index;
2701 int index_max = self->audio_index + 1;
2702 if ( self->audio_index == INT_MAX )
2703 {
2704 index = 0;
2705 index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
2706 *channels = self->total_channels;
2707 *samples = mlt_audio_calculate_frame_samples( fps, self->max_frequency, position );
2708 *frequency = self->max_frequency;
2709 }
2710
2711 // Initialize the buffers
2712 for ( ; index < index_max && index < MAX_AUDIO_STREAMS; index++ )
2713 {
2714 // Get codec context
2715 AVCodecContext *codec_context = self->audio_codec[ index ];
2716
2717 if ( codec_context && !self->audio_buffer[ index ] )
2718 {
2719 if ( self->audio_index != INT_MAX && !mlt_properties_get( MLT_PRODUCER_PROPERTIES(self->parent), "request_channel_layout" ) )
2720 codec_context->request_channel_layout = av_get_default_channel_layout( *channels );
2721 sizeof_sample = sample_bytes( codec_context );
2722
2723 // Check for audio buffer and create if necessary
2724 self->audio_buffer_size[ index ] = MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2725 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2726
2727 // Check for decoder buffer and create if necessary
2728 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2729 }
2730 }
2731
2732 // Get the audio if required
2733 if ( !paused && *frequency > 0 )
2734 {
2735 int ret = 0;
2736 int got_audio = 0;
2737 AVPacket pkt;
2738 mlt_channel_layout layout = mlt_channel_auto;
2739
2740 av_init_packet( &pkt );
2741
2742 // Caller requested number samples based on requested sample rate.
2743 if ( self->audio_index != INT_MAX )
2744 *samples = mlt_audio_calculate_frame_samples( fps, self->audio_codec[ self->audio_index ]->sample_rate, position );
2745
2746 while ( ret >= 0 && !got_audio )
2747 {
2748 // Check if the buffer already contains the samples required
2749 if ( self->audio_index != INT_MAX &&
2750 self->audio_used[ self->audio_index ] >= *samples &&
2751 ignore[ self->audio_index ] == 0 )
2752 {
2753 got_audio = 1;
2754 break;
2755 }
2756 else if ( self->audio_index == INT_MAX )
2757 {
2758 // Check if there is enough audio for all streams
2759 got_audio = 1;
2760 for ( index = 0; got_audio && index < index_max; index++ )
2761 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2762 got_audio = 0;
2763 if ( got_audio )
2764 break;
2765 }
2766
2767 // Read a packet
2768 pthread_mutex_lock( &self->packets_mutex );
2769 if ( mlt_deque_count( self->apackets ) )
2770 {
2771 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2772 av_packet_ref( &pkt, tmp );
2773 av_packet_free( &tmp );
2774 }
2775 else
2776 {
2777 ret = av_read_frame( context, &pkt );
2778 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2779 {
2780 mlt_deque_push_back( self->vpackets, av_packet_clone(&pkt) );
2781 }
2782 else if ( ret < 0 )
2783 {
2784 mlt_producer producer = self->parent;
2785 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2786 if ( ret != AVERROR_EOF )
2787 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "av_read_frame returned error %d inside get_audio\n", ret );
2788 if ( !self->seekable && mlt_properties_get_int( properties, "reconnect" ) )
2789 {
2790 // Try to reconnect to live sources by closing context and codecs,
2791 // and letting next call to get_frame() reopen.
2792 prepare_reopen( self );
2793 pthread_mutex_unlock( &self->packets_mutex );
2794 goto exit_get_audio;
2795 }
2796 if ( !self->seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) )
2797 {
2798 mlt_log_fatal( MLT_PRODUCER_SERVICE(producer), "Exiting with error due to disconnected source.\n" );
2799 exit( EXIT_FAILURE );
2800 }
2801 }
2802 }
2803 pthread_mutex_unlock( &self->packets_mutex );
2804
2805 // We only deal with audio from the selected audio index
2806 index = pkt.stream_index;
2807 if ( index < MAX_AUDIO_STREAMS && ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2808 ( self->audio_index == INT_MAX && context->streams[ index ]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO ) ) )
2809 {
2810 ret = decode_audio( self, &ignore[index], &pkt, *samples, real_timecode, fps );
2811 }
2812
2813 if ( self->seekable || index != self->video_index )
2814 av_packet_unref( &pkt );
2815 }
2816 self->is_audio_synchronizing = 0;
2817
2818 // Set some additional return values
2819 *format = mlt_audio_s16;
2820 if ( self->audio_index != INT_MAX )
2821 {
2822 index = self->audio_index;
2823 *channels = self->audio_codec[ index ]->channels;
2824 *frequency = self->audio_codec[ index ]->sample_rate;
2825 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2826 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2827 if( self->audio_codec[ index ]->channel_layout == 0 )
2828 layout = av_channel_layout_to_mlt( av_get_default_channel_layout( self->audio_codec[ index ]->channels ) );
2829 else
2830 layout = av_channel_layout_to_mlt( self->audio_codec[ index ]->channel_layout );
2831 }
2832 else if ( self->audio_index == INT_MAX )
2833 {
2834 layout = mlt_channel_independent;
2835 for ( index = 0; index < index_max; index++ )
2836 if ( self->audio_codec[ index ] )
2837 {
2838 // XXX: This only works if all audio tracks have the same sample format.
2839 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2840 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2841 break;
2842 }
2843 }
2844 mlt_properties_set( MLT_FRAME_PROPERTIES(frame), "channel_layout", mlt_audio_channel_layout_name( layout ) );
2845
2846 // Allocate and set the frame's audio buffer
2847 int size = mlt_audio_format_size( *format, *samples, *channels );
2848 *buffer = mlt_pool_alloc( size );
2849 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2850
2851 // Interleave tracks if audio_index=all
2852 if ( self->audio_index == INT_MAX )
2853 {
2854 uint8_t *dest = *buffer;
2855 int i;
2856 for ( i = 0; i < *samples; i++ )
2857 {
2858 for ( index = 0; index < index_max; index++ )
2859 if ( self->audio_codec[ index ] )
2860 {
2861 int current_channels = self->audio_codec[ index ]->channels;
2862 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2863 memcpy( dest, src, current_channels * sizeof_sample );
2864 dest += current_channels * sizeof_sample;
2865 }
2866 }
2867 for ( index = 0; index < index_max; index++ )
2868 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2869 {
2870 int current_channels = self->audio_codec[ index ]->channels;
2871 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2872 self->audio_used[index] -= *samples;
2873 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2874 }
2875 }
2876 // Copy a single track to the output buffer
2877 else
2878 {
2879 index = self->audio_index;
2880 uint8_t silence = *format == mlt_audio_u8 ? 0x80 : 0;
2881
2882 // Now handle the audio if we have enough
2883 if ( self->audio_used[ index ] > 0 )
2884 {
2885 uint8_t *src = self->audio_buffer[ index ];
2886 // copy samples from audio_buffer
2887 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2888 memcpy( *buffer, src, size * *channels * sizeof_sample );
2889 // supply the remaining requested samples as silence
2890 if ( *samples > self->audio_used[ index ] )
2891 memset( *buffer + size * *channels * sizeof_sample, silence, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2892 // reposition the samples within audio_buffer
2893 self->audio_used[ index ] -= size;
2894 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2895 }
2896 else
2897 {
2898 // Otherwise fill with silence
2899 memset( *buffer, silence, *samples * *channels * sizeof_sample );
2900 }
2901 }
2902 }
2903 else
2904 {
2905 exit_get_audio:
2906 // Get silence and don't touch the context
2907 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2908 }
2909
2910 // Regardless of speed (other than paused), we expect to get the next frame
2911 if ( !paused )
2912 self->audio_expected = position + 1;
2913
2914 pthread_mutex_unlock( &self->audio_mutex );
2915
2916 return 0;
2917 }
2918
2919 /** Initialize the audio codec context.
2920 */
2921
audio_codec_init(producer_avformat self,int index,mlt_properties properties)2922 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2923 {
2924 // Initialise the codec if necessary
2925 if ( !self->audio_codec[ index ] )
2926 {
2927 // Get codec context
2928 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2929
2930 // Find the codec
2931 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2932 if ( mlt_properties_get( properties, "acodec" ) )
2933 {
2934 if ( !( codec = avcodec_find_decoder_by_name( mlt_properties_get( properties, "acodec" ) ) ) )
2935 codec = avcodec_find_decoder( codec_context->codec_id );
2936 }
2937
2938 // If we don't have a codec and we can't initialise it, we can't do much more...
2939 pthread_mutex_lock( &self->open_mutex );
2940 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2941 {
2942 // Now store the codec with its destructor
2943 if ( self->audio_codec[ index ] )
2944 avcodec_close( self->audio_codec[ index ] );
2945 self->audio_codec[ index ] = codec_context;
2946 self->audio_index = index;
2947 }
2948 else
2949 {
2950 // Remember that we can't use self later
2951 self->audio_index = -1;
2952 }
2953 pthread_mutex_unlock( &self->open_mutex );
2954
2955 // Process properties as AVOptions
2956 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2957 if ( codec && codec->priv_class && codec_context->priv_data )
2958 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2959 }
2960 return self->audio_codec[ index ] && self->audio_index > -1;
2961 }
2962
2963 /** Set up audio handling.
2964 */
2965
producer_set_up_audio(producer_avformat self,mlt_frame frame)2966 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2967 {
2968 // Get the producer
2969 mlt_producer producer = self->parent;
2970
2971 // Get the properties
2972 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2973
2974 // Fetch the audio format context
2975 AVFormatContext *context = self->audio_format;
2976
2977 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2978
2979 // Get the audio_index
2980 int index = mlt_properties_get_int( properties, "audio_index" );
2981
2982 // Handle all audio tracks
2983 if ( self->audio_index > -1 &&
2984 mlt_properties_get( properties, "audio_index" ) &&
2985 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2986 index = INT_MAX;
2987
2988 // Reopen the file if necessary
2989 if ( !context && self->audio_index > -1 && index > -1 )
2990 {
2991 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2992 mlt_properties_get( properties, "resource" ), 1, 0 );
2993 context = self->audio_format;
2994 }
2995
2996 // Exception handling for audio_index
2997 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
2998 {
2999 for ( index = context->nb_streams - 1;
3000 index >= 0 && context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_AUDIO;
3001 index-- );
3002 mlt_properties_set_int( properties, "audio_index", index );
3003 }
3004 if ( context && index > -1 && index < INT_MAX &&
3005 context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_AUDIO )
3006 {
3007 index = self->audio_index;
3008 mlt_properties_set_int( properties, "audio_index", index );
3009 }
3010 if ( context && index > -1 && index < INT_MAX &&
3011 pick_audio_format( context->streams[ index ]->codecpar->format ) == mlt_audio_none )
3012 {
3013 index = -1;
3014 }
3015
3016 // Update the audio properties if the index changed
3017 if ( context && self->audio_index > -1 && index != self->audio_index )
3018 {
3019 pthread_mutex_lock( &self->open_mutex );
3020 unsigned i = 0;
3021 int index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
3022 for (i = 0; i < index_max; i++) {
3023 if (self->audio_codec[i]) {
3024 avcodec_close(self->audio_codec[i]);
3025 self->audio_codec[i] = NULL;
3026 }
3027 }
3028 pthread_mutex_unlock( &self->open_mutex );
3029 }
3030
3031 // Get the codec(s)
3032 if ( context && index == INT_MAX )
3033 {
3034 unsigned int index;
3035 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
3036 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
3037 for ( index = 0; index < context->nb_streams && index < MAX_AUDIO_STREAMS; index++ )
3038 {
3039 if ( context->streams[ index ]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO )
3040 audio_codec_init( self, index, properties );
3041 }
3042 self->audio_index = INT_MAX;
3043 }
3044 else if ( context && index > -1 && index < MAX_AUDIO_STREAMS &&
3045 audio_codec_init( self, index, properties ) )
3046 {
3047 mlt_properties_set_int( frame_properties, "audio_frequency", self->audio_codec[ index ]->sample_rate );
3048 mlt_properties_set_int( frame_properties, "audio_channels", self->audio_codec[ index ]->channels );
3049 }
3050 if ( context && index > -1 )
3051 {
3052 // Add our audio operation
3053 mlt_frame_push_audio( frame, self );
3054 mlt_frame_push_audio( frame, producer_get_audio );
3055 }
3056 }
3057
3058 /** Our get frame implementation.
3059 */
3060
producer_get_frame(mlt_producer producer,mlt_frame_ptr frame,int index)3061 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
3062 {
3063 // Access the private data
3064 (void) index; // unused
3065 mlt_service service = MLT_PRODUCER_SERVICE( producer );
3066 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
3067 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
3068
3069 // If cache miss
3070 if ( !self )
3071 {
3072 self = calloc( 1, sizeof( struct producer_avformat_s ) );
3073 producer->child = self;
3074 self->parent = producer;
3075 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
3076 cache_item = mlt_service_cache_get( service, "producer_avformat" );
3077 }
3078
3079 // Create an empty frame
3080 *frame = mlt_frame_init( service);
3081
3082 if ( *frame )
3083 {
3084 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
3085 }
3086 else
3087 {
3088 mlt_cache_item_close( cache_item );
3089 return 1;
3090 }
3091
3092 // Update timecode on the frame we're creating
3093 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
3094
3095 // Set up the video
3096 producer_set_up_video( self, *frame );
3097
3098 // Set up the audio
3099 producer_set_up_audio( self, *frame );
3100
3101 // Set the position of this producer
3102 mlt_position position = mlt_producer_frame( producer );
3103 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "original_position", position );
3104
3105 // Calculate the next timecode
3106 mlt_producer_prepare_next( producer );
3107
3108 return 0;
3109 }
3110
producer_avformat_close(producer_avformat self)3111 static void producer_avformat_close( producer_avformat self )
3112 {
3113 mlt_log_debug( NULL, "producer_avformat_close\n" );
3114
3115 // Cleanup av contexts
3116 av_packet_unref( &self->pkt );
3117 av_frame_free( &self->video_frame );
3118 av_frame_free( &self->audio_frame );
3119
3120 #if USE_HWACCEL
3121 av_buffer_unref( &self->hwaccel.device_ctx );
3122 #endif
3123
3124 if ( self->is_mutex_init )
3125 pthread_mutex_lock( &self->open_mutex );
3126 int i;
3127 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
3128 {
3129 mlt_pool_release( self->audio_buffer[i] );
3130 av_free( self->decode_buffer[i] );
3131 if ( self->audio_codec[i] )
3132 avcodec_close( self->audio_codec[i] );
3133 self->audio_codec[i] = NULL;
3134 }
3135 if ( self->video_codec )
3136 avcodec_close( self->video_codec );
3137 self->video_codec = NULL;
3138 // Close the file
3139 if ( self->dummy_context )
3140 avformat_close_input( &self->dummy_context );
3141 if ( self->seekable && self->audio_format )
3142 avformat_close_input( &self->audio_format );
3143 if ( self->video_format )
3144 avformat_close_input( &self->video_format );
3145 if ( self->is_mutex_init )
3146 pthread_mutex_unlock( &self->open_mutex );
3147 #ifdef AVFILTER
3148 avfilter_graph_free(&self->vfilter_graph);
3149 #endif
3150
3151 // Cleanup caches.
3152 mlt_cache_close( self->image_cache );
3153 if ( self->last_good_frame )
3154 mlt_frame_close( self->last_good_frame );
3155
3156 // Cleanup the mutexes
3157 if ( self->is_mutex_init )
3158 {
3159 pthread_mutex_destroy( &self->audio_mutex );
3160 pthread_mutex_destroy( &self->video_mutex );
3161 pthread_mutex_destroy( &self->packets_mutex );
3162 pthread_mutex_destroy( &self->open_mutex );
3163 }
3164
3165 // Cleanup the packet queues
3166 AVPacket *pkt;
3167 if ( self->apackets )
3168 {
3169 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
3170 {
3171 av_packet_free( &pkt );
3172 }
3173 mlt_deque_close( self->apackets );
3174 self->apackets = NULL;
3175 }
3176 if ( self->vpackets )
3177 {
3178 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
3179 {
3180 av_packet_free( &pkt );
3181 }
3182 mlt_deque_close( self->vpackets );
3183 self->vpackets = NULL;
3184 }
3185
3186 free( self );
3187 }
3188
producer_close(mlt_producer parent)3189 static void producer_close( mlt_producer parent )
3190 {
3191 // Remove this instance from the cache
3192 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
3193
3194 // Close the parent
3195 parent->close = NULL;
3196 mlt_producer_close( parent );
3197
3198 // Free the memory
3199 free( parent );
3200 }
3201