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 &&
402 (codec_params->codec_id != AV_CODEC_ID_MJPEG ||
403 codec_context->time_base.num != 1 ||
404 codec_context->time_base.den != 90000)) {
405 self->video_index = i;
406 }
407 mlt_properties_set( meta_media, key, "video" );
408 snprintf( key, sizeof(key), "meta.media.%u.stream.frame_rate", i );
409 double ffmpeg_fps = av_q2d( context->streams[ i ]->avg_frame_rate );
410 mlt_properties_set_double( meta_media, key, ffmpeg_fps );
411
412 #if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(56, 1, 0)
413 const char *projection = get_projection(context->streams[i]);
414 if (projection) {
415 snprintf(key, sizeof(key), "meta.media.%u.stream.projection", i);
416 mlt_properties_set_string(meta_media, key, projection);
417 }
418 #endif
419 snprintf( key, sizeof(key), "meta.media.%u.stream.sample_aspect_ratio", i );
420 mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->sample_aspect_ratio ) );
421 snprintf( key, sizeof(key), "meta.media.%u.codec.width", i );
422 mlt_properties_set_int( meta_media, key, codec_params->width );
423 snprintf( key, sizeof(key), "meta.media.%u.codec.height", i );
424 mlt_properties_set_int( meta_media, key, codec_params->height );
425 snprintf( key, sizeof(key), "meta.media.%u.codec.rotate", i );
426 mlt_properties_set_int( meta_media, key, get_rotation(context->streams[i]) );
427 snprintf( key, sizeof(key), "meta.media.%u.codec.frame_rate", i );
428 AVRational frame_rate = { codec_context->time_base.den, codec_context->time_base.num * codec_context->ticks_per_frame };
429 mlt_properties_set_double( meta_media, key, av_q2d( frame_rate ) );
430 snprintf( key, sizeof(key), "meta.media.%u.codec.pix_fmt", i );
431 mlt_properties_set( meta_media, key, av_get_pix_fmt_name( codec_params->format ) );
432 snprintf( key, sizeof(key), "meta.media.%u.codec.sample_aspect_ratio", i );
433 mlt_properties_set_double( meta_media, key, av_q2d( codec_params->sample_aspect_ratio ) );
434 snprintf( key, sizeof(key), "meta.media.%u.codec.colorspace", i );
435 switch ( codec_params->color_space )
436 {
437 case AVCOL_SPC_SMPTE240M:
438 mlt_properties_set_int( meta_media, key, 240 );
439 break;
440 case AVCOL_SPC_BT470BG:
441 case AVCOL_SPC_SMPTE170M:
442 mlt_properties_set_int( meta_media, key, 601 );
443 break;
444 case AVCOL_SPC_BT709:
445 mlt_properties_set_int( meta_media, key, 709 );
446 break;
447 case AVCOL_SPC_UNSPECIFIED:
448 case AVCOL_SPC_RESERVED:
449 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
450 mlt_properties_set_int( meta_media, key, codec_params->width * codec_params->height > 750000 ? 709 : 601 );
451 break;
452 default:
453 mlt_properties_set_int( meta_media, key, codec_context->colorspace );
454 break;
455 }
456 if ( codec_params->color_trc && codec_params->color_trc != AVCOL_TRC_UNSPECIFIED )
457 {
458 snprintf( key, sizeof(key), "meta.media.%u.codec.color_trc", i );
459 mlt_properties_set_double( meta_media, key, codec_params->color_trc );
460 }
461 break;
462 case AVMEDIA_TYPE_AUDIO:
463 if ( !codec_params->channels )
464 break;
465 // Use first audio stream
466 if ( self->audio_index < 0 && pick_audio_format( codec_params->format ) != mlt_audio_none )
467 self->audio_index = i;
468
469 mlt_properties_set( meta_media, key, "audio" );
470 snprintf( key, sizeof(key), "meta.media.%u.codec.sample_fmt", i );
471 mlt_properties_set( meta_media, key, av_get_sample_fmt_name( codec_params->format ) );
472 snprintf( key, sizeof(key), "meta.media.%u.codec.sample_rate", i );
473 mlt_properties_set_int( meta_media, key, codec_params->sample_rate );
474 snprintf( key, sizeof(key), "meta.media.%u.codec.channels", i );
475 mlt_properties_set_int( meta_media, key, codec_params->channels );
476 break;
477 default:
478 break;
479 }
480 // snprintf( key, sizeof(key), "meta.media.%u.stream.time_base", i );
481 // mlt_properties_set_double( meta_media, key, av_q2d( context->streams[ i ]->time_base ) );
482 snprintf( key, sizeof(key), "meta.media.%u.codec.name", i );
483 mlt_properties_set( meta_media, key, codec->name );
484 snprintf( key, sizeof(key), "meta.media.%u.codec.long_name", i );
485 mlt_properties_set( meta_media, key, codec->long_name );
486 snprintf( key, sizeof(key), "meta.media.%u.codec.bit_rate", i );
487 mlt_properties_set_int64( meta_media, key, codec_params->bit_rate );
488 // snprintf( key, sizeof(key), "meta.media.%u.codec.time_base", i );
489 // mlt_properties_set_double( meta_media, key, av_q2d( codec_context->time_base ) );
490 // snprintf( key, sizeof(key), "meta.media.%u.codec.profile", i );
491 // mlt_properties_set_int( meta_media, key, codec_context->profile );
492 // snprintf( key, sizeof(key), "meta.media.%u.codec.level", i );
493 // mlt_properties_set_int( meta_media, key, codec_context->level );
494
495 // Read Metadata
496 while ( ( tag = av_dict_get( stream->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
497 {
498 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
499 {
500 snprintf( key, sizeof(key), "meta.attr.%u.stream.%s.markup", i, tag->key );
501 char* value = filter_restricted( tag->value );
502 mlt_properties_set( meta_media, key, value );
503 free( value );
504 }
505 }
506 }
507
508 // Use the album art if that is all we have
509 if (self->video_index < 0 && first_video_index >= 0)
510 self->video_index = first_video_index;
511
512 while ( ( tag = av_dict_get( context->metadata, "", tag, AV_DICT_IGNORE_SUFFIX ) ) )
513 {
514 if ( tag->value && strcmp( tag->value, "" ) && strcmp( tag->value, "und" ) )
515 {
516 snprintf( key, sizeof(key), "meta.attr.%s.markup", tag->key );
517 char* value = filter_restricted( tag->value );
518 mlt_properties_set( meta_media, key, value );
519 free( value );
520 }
521 }
522
523 return meta_media;
524 }
525
get_aspect_ratio(mlt_properties properties,AVStream * stream,AVCodecParameters * codec_params)526 static void get_aspect_ratio( mlt_properties properties, AVStream *stream, AVCodecParameters *codec_params )
527 {
528 AVRational sar = stream->sample_aspect_ratio;
529 if ( sar.num <= 0 || sar.den <= 0 )
530 sar = codec_params->sample_aspect_ratio;
531 if ( sar.num <= 0 || sar.den <= 0 )
532 sar.num = sar.den = 1;
533 mlt_properties_set_int( properties, "meta.media.sample_aspect_num", sar.num );
534 mlt_properties_set_int( properties, "meta.media.sample_aspect_den", sar.den );
535 mlt_properties_set_double( properties, "aspect_ratio", av_q2d( sar ) );
536 }
537
parse_url(mlt_profile profile,const char * URL,AVInputFormat ** format,AVDictionary ** params)538 static char* parse_url( mlt_profile profile, const char* URL, AVInputFormat **format, AVDictionary **params )
539 {
540 (void) profile; // unused
541 if ( !URL ) return NULL;
542
543 char *protocol = strdup( URL );
544 char *url = strchr( protocol, ':' );
545
546 // Truncate protocol string
547 if (url && (url - protocol) > 1 && avio_check(URL, 0) < 0) { // if defined and not a drive letter
548 url[0] = '\0';
549 ++url;
550 mlt_log_debug( NULL, "%s: protocol=%s resource=%s\n", __FUNCTION__, protocol, url );
551
552 // Lookup the format
553 *format = av_find_input_format( protocol );
554 } else {
555 url = protocol;
556 }
557
558 // Eat the format designator
559 char *result = url;
560
561 // support for legacy width and height parameters
562 char *width = NULL;
563 char *height = NULL;
564
565 // Parse out params
566 char* query = strchr( url, '?' );
567 if (*format) {
568 // Query string delimiter is '?'
569 url = ( query && query > url && query[-1] != '\\' ) ? query : NULL;
570 } else {
571 // Ignore unescaped question marks
572 while ( query && query > url && query[-1] != '\\' ) {
573 query = strchr( query + 1, '?' );
574 }
575 // Query string delimiter is '\?'
576 url = ( query && query > url && query[-1] == '\\' ) ? query : NULL;
577 if (url) url[-1] = '\0'; // null the backslash
578 }
579 while ( url )
580 {
581 url[0] = '\0';
582 char *name = strdup( ++url );
583 char *value = strchr( name, '=' );
584 if ( !value )
585 // Also accept : as delimiter for backwards compatibility.
586 value = strchr( name, ':' );
587 if ( value )
588 {
589 value[0] = '\0';
590 value++;
591 char *t = strchr( value, '&' );
592 if ( t )
593 t[0] = 0;
594 // translate old parameters to new av_dict names
595 if ( !strcmp( name, "frame_rate" ) )
596 av_dict_set( params, "framerate", value, 0 );
597 else if ( !strcmp( name, "pix_fmt" ) )
598 av_dict_set( params, "pixel_format", value, 0 );
599 else if ( !strcmp( name, "width" ) )
600 width = strdup( value );
601 else if ( !strcmp( name, "height" ) )
602 height = strdup( value );
603 else
604 // generic demux/device option support
605 av_dict_set( params, name, value, 0 );
606 }
607 free( name );
608 url = strchr( url, '&' );
609 }
610 // continued support for legacy width and height parameters
611 if ( width && height )
612 {
613 char *s = malloc( strlen( width ) + strlen( height ) + 2 );
614 strcpy( s, width );
615 strcat( s, "x");
616 strcat( s, height );
617 av_dict_set( params, "video_size", s, 0 );
618 free( s );
619 }
620 free( width );
621 free( height );
622
623 result = strdup(result);
624 free( protocol );
625 mlt_log_debug(NULL, "[producer avformat] %s filename = %s\n", __FUNCTION__, result);
626 return result;
627 }
628
pick_pix_fmt(enum AVPixelFormat pix_fmt)629 static enum AVPixelFormat pick_pix_fmt( enum AVPixelFormat pix_fmt )
630 {
631 switch ( pix_fmt )
632 {
633 case AV_PIX_FMT_ARGB:
634 case AV_PIX_FMT_RGBA:
635 case AV_PIX_FMT_ABGR:
636 case AV_PIX_FMT_BGRA:
637 return AV_PIX_FMT_RGBA;
638 #if defined(FFUDIV)
639 case AV_PIX_FMT_BAYER_RGGB16LE:
640 return AV_PIX_FMT_RGB24;
641 #endif
642 #if USE_HWACCEL
643 case AV_PIX_FMT_VAAPI:
644 case AV_PIX_FMT_CUDA:
645 case AV_PIX_FMT_VIDEOTOOLBOX:
646 case AV_PIX_FMT_DXVA2_VLD:
647 case AV_PIX_FMT_D3D11:
648 return AV_PIX_FMT_YUV420P;
649 #endif
650 default:
651 return AV_PIX_FMT_YUV422P;
652 }
653 }
654
get_basic_info(producer_avformat self,mlt_profile profile,const char * filename)655 static int get_basic_info( producer_avformat self, mlt_profile profile, const char *filename )
656 {
657 int error = 0;
658
659 // Get the properties
660 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
661
662 AVFormatContext *format = self->video_format;
663
664 // Get the duration
665 if ( mlt_properties_get_position( properties, "length" ) <= 0 ||
666 mlt_properties_get_position( properties, "out" ) <= 0 )
667 {
668 if ( format->duration != AV_NOPTS_VALUE )
669 {
670 // This isn't going to be accurate for all formats
671 // We will treat everything with the producer fps.
672 mlt_position frames = ( mlt_position ) lrint( format->duration * mlt_profile_fps( profile ) / AV_TIME_BASE );
673 if ( mlt_properties_get_position( properties, "out" ) <= 0 )
674 mlt_properties_set_position( properties, "out", frames - 1 );
675 if ( mlt_properties_get_position( properties, "length" ) <= 0 )
676 mlt_properties_set_position( properties, "length", frames );
677 }
678 else if ( format->nb_streams > 0 && format->streams[0]->codecpar && format->streams[0]->codecpar->codec_id == AV_CODEC_ID_WEBP )
679 {
680 char *e = getenv( "MLT_DEFAULT_PRODUCER_LENGTH" );
681 int p = e ? atoi( e ) : 15000;
682 mlt_properties_set_int( properties, "out", MAX(0, p - 1) );
683 mlt_properties_set_int( properties, "length", p );
684 }
685 else
686 {
687 // Set live sources to run forever
688 if ( mlt_properties_get_position( properties, "length" ) <= 0 )
689 mlt_properties_set_position( properties, "length", INT_MAX );
690 if ( mlt_properties_get_position( properties, "out" ) <= 0 )
691 mlt_properties_set_position( properties, "out", INT_MAX - 1 );
692 mlt_properties_set( properties, "eof", "loop" );
693 }
694 }
695
696 // Check if we're seekable
697 // avdevices are typically AVFMT_NOFILE and not seekable
698 self->seekable = !format->iformat || !( format->iformat->flags & AVFMT_NOFILE );
699 if ( format->pb )
700 {
701 // protocols can indicate if they support seeking
702 self->seekable = format->pb->seekable;
703 }
704 if ( self->seekable )
705 {
706 // Do a more rigorous test of seekable on a disposable context
707 if ( format->nb_streams > 0 && format->streams[0]->codecpar && format->streams[0]->codecpar->codec_id != AV_CODEC_ID_WEBP )
708 self->seekable = av_seek_frame( format, -1, format->start_time, AVSEEK_FLAG_BACKWARD ) >= 0;
709 mlt_properties_set_int( properties, "seekable", self->seekable );
710 self->dummy_context = format;
711 self->video_format = NULL;
712 avformat_open_input( &self->video_format, filename, NULL, NULL );
713 avformat_find_stream_info( self->video_format, NULL );
714 format = self->video_format;
715 }
716 self->video_seekable = self->seekable;
717
718 // Fetch the width, height and aspect ratio
719 if ( self->video_index != -1 )
720 {
721 AVCodecParameters *codec_params = format->streams[ self->video_index ]->codecpar;
722 mlt_properties_set_int( properties, "width", codec_params->width );
723 mlt_properties_set_int( properties, "height", codec_params->height );
724 get_aspect_ratio( properties, format->streams[ self->video_index ], codec_params );
725
726 int pix_fmt = codec_params->format;
727 pick_av_pixel_format( &pix_fmt );
728 if ( pix_fmt != AV_PIX_FMT_NONE ) {
729 // Verify that we can convert this to one of our image formats.
730 struct SwsContext *context = sws_getContext( codec_params->width, codec_params->height, pix_fmt,
731 codec_params->width, codec_params->height, pick_pix_fmt( codec_params->format ), SWS_BILINEAR, NULL, NULL, NULL);
732 if ( context )
733 sws_freeContext( context );
734 else
735 error = 1;
736 } else {
737 self->video_index = -1;
738 }
739 }
740 return error;
741 }
742
743 #ifdef AVFILTER
setup_video_filters(producer_avformat self)744 static int setup_video_filters( producer_avformat self )
745 {
746 mlt_properties properties = MLT_PRODUCER_PROPERTIES(self->parent);
747 AVFormatContext *format = self->video_format;
748 AVStream* stream = format->streams[ self->video_index ];
749 AVCodecParameters *codec_params = stream->codecpar;
750
751 self->vfilter_graph = avfilter_graph_alloc();
752
753 // From ffplay.c:configure_video_filters().
754 char buffersrc_args[256];
755 snprintf(buffersrc_args, sizeof(buffersrc_args),
756 "video_size=%dx%d:pix_fmt=%d:time_base=%d/%d:pixel_aspect=%d/%d:frame_rate=%d/%d",
757 codec_params->width, codec_params->height, codec_params->format,
758 stream->time_base.num, stream->time_base.den,
759 mlt_properties_get_int(properties, "meta.media.sample_aspect_num"),
760 FFMAX(mlt_properties_get_int(properties, "meta.media.sample_aspect_den"), 1),
761 stream->avg_frame_rate.num, FFMAX(stream->avg_frame_rate.den, 1));
762
763 int result = avfilter_graph_create_filter(&self->vfilter_in, avfilter_get_by_name("buffer"),
764 "mlt_buffer", buffersrc_args, NULL, self->vfilter_graph);
765
766 if (result >= 0) {
767 result = avfilter_graph_create_filter(&self->vfilter_out, avfilter_get_by_name("buffersink"),
768 "mlt_buffersink", NULL, NULL, self->vfilter_graph);
769
770 if (result >= 0) {
771 enum AVPixelFormat pix_fmts[] = { codec_params->format, AV_PIX_FMT_NONE };
772 result = av_opt_set_int_list(self->vfilter_out, "pix_fmts", pix_fmts, AV_PIX_FMT_NONE, AV_OPT_SEARCH_CHILDREN);
773 }
774 }
775
776 return result;
777 }
778
insert_filter(AVFilterGraph * graph,AVFilterContext ** last_filter,const char * name,const char * args)779 static int insert_filter(AVFilterGraph *graph, AVFilterContext **last_filter, const char *name, const char *args)
780 {
781 AVFilterContext *filt_ctx;
782 int result = avfilter_graph_create_filter(&filt_ctx, avfilter_get_by_name(name),
783 name, args, NULL, graph);
784 if (result >= 0) {
785 result = avfilter_link(filt_ctx, 0, *last_filter, 0);
786 if (result >= 0)
787 *last_filter = filt_ctx;
788 }
789 return result;
790 }
791 #endif
792
793 /** Open the file.
794 */
795
producer_open(producer_avformat self,mlt_profile profile,const char * URL,int take_lock,int test_open)796 static int producer_open(producer_avformat self, mlt_profile profile, const char *URL, int take_lock, int test_open )
797 {
798 // Return an error code (0 == no error)
799 int error = 0;
800 mlt_properties properties = MLT_PRODUCER_PROPERTIES( self->parent );
801
802 if ( !self->is_mutex_init )
803 {
804 pthread_mutexattr_t attr;
805 pthread_mutexattr_init(&attr);
806 pthread_mutexattr_settype(&attr, PTHREAD_MUTEX_RECURSIVE);
807 pthread_mutex_init( &self->audio_mutex, &attr );
808 pthread_mutex_init( &self->video_mutex, &attr );
809 pthread_mutex_init( &self->packets_mutex, &attr );
810 pthread_mutex_init( &self->open_mutex, &attr );
811 self->is_mutex_init = 1;
812 }
813
814 // Lock the service
815 if ( take_lock )
816 {
817 pthread_mutex_lock( &self->audio_mutex );
818 pthread_mutex_lock( &self->video_mutex );
819 }
820 mlt_events_block( properties, self->parent );
821
822 // Parse URL
823 AVInputFormat *format = NULL;
824 AVDictionary *params = NULL;
825 char *filename = parse_url( profile, URL, &format, ¶ms );
826
827 // Now attempt to open the file or device with filename
828 error = avformat_open_input( &self->video_format, filename, format, ¶ms ) < 0;
829 if ( error )
830 // If the URL is a network stream URL, then we probably need to open with full URL
831 error = avformat_open_input( &self->video_format, URL, format, ¶ms ) < 0;
832
833 // Set MLT properties onto video AVFormatContext
834 if ( !error && self->video_format )
835 {
836 apply_properties( self->video_format, properties, AV_OPT_FLAG_DECODING_PARAM );
837 if ( self->video_format->iformat && self->video_format->iformat->priv_class && self->video_format->priv_data )
838 apply_properties( self->video_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
839 }
840
841 // If successful, then try to get additional info
842 if ( !error && self->video_format )
843 {
844 // Get the stream info
845 error = avformat_find_stream_info( self->video_format, NULL ) < 0;
846
847 // Continue if no error
848 if ( !error && self->video_format )
849 {
850 // Find default audio and video streams
851 find_default_streams( self );
852 error = get_basic_info( self, profile, filename );
853
854 // Initialize position info
855 self->first_pts = AV_NOPTS_VALUE;
856 self->last_position = POSITION_INITIAL;
857
858 #if USE_HWACCEL
859 AVDictionaryEntry *hwaccel = av_dict_get( params, "hwaccel", NULL, 0 );
860 AVDictionaryEntry *hwaccel_device = av_dict_get( params, "hwaccel_device", NULL, 0 );
861
862 if ( hwaccel && hwaccel->value )
863 {
864 // Leaving `device=NULL` will cause query string parameter `hwaccel_device` to be ignored
865 char *device = NULL;
866 if ( !strcmp( hwaccel->value, "vaapi" ) )
867 {
868 self->hwaccel.pix_fmt = AV_PIX_FMT_VAAPI;
869 self->hwaccel.device_type = AV_HWDEVICE_TYPE_VAAPI;
870 device = "/dev/dri/renderD128";
871 }
872 else if ( !strcmp( hwaccel->value, "cuda" ) )
873 {
874 self->hwaccel.pix_fmt = AV_PIX_FMT_CUDA;
875 self->hwaccel.device_type = AV_HWDEVICE_TYPE_CUDA;
876 device = "0";
877 }
878 else if ( !strcmp( hwaccel->value, "videotoolbox" ) )
879 {
880 self->hwaccel.pix_fmt = AV_PIX_FMT_VIDEOTOOLBOX;
881 self->hwaccel.device_type = AV_HWDEVICE_TYPE_VIDEOTOOLBOX;
882 }
883 else if ( !strcmp( hwaccel->value, "d3d11va" ) )
884 {
885 self->hwaccel.pix_fmt = AV_PIX_FMT_D3D11;
886 self->hwaccel.device_type = AV_HWDEVICE_TYPE_D3D11VA;
887 device = "0";
888 }
889 else if ( !strcmp( hwaccel->value, "dxva2" ) )
890 {
891 self->hwaccel.pix_fmt = AV_PIX_FMT_DXVA2_VLD;
892 self->hwaccel.device_type = AV_HWDEVICE_TYPE_DXVA2;
893 device = "0";
894 }
895 else
896 {
897 // TODO: init other hardware types
898 }
899
900 if (device) {
901 if (hwaccel_device && hwaccel_device->value)
902 device = hwaccel_device->value;
903 memcpy(self->hwaccel.device, device, strlen(device));
904 }
905 }
906 #endif
907
908 if ( !self->audio_format )
909 {
910 // We're going to cheat here - for seekable A/V files, we will have separate contexts
911 // to support independent seeking of audio from video.
912 // TODO: Is this really necessary?
913 if ( self->audio_index != -1 && self->video_index != -1 )
914 {
915 if ( self->seekable )
916 {
917 // And open again for our audio context
918 avformat_open_input( &self->audio_format, filename, NULL, NULL );
919 apply_properties( self->audio_format, properties, AV_OPT_FLAG_DECODING_PARAM );
920 if ( self->audio_format->iformat && self->audio_format->iformat->priv_class && self->audio_format->priv_data )
921 apply_properties( self->audio_format->priv_data, properties, AV_OPT_FLAG_DECODING_PARAM );
922 avformat_find_stream_info( self->audio_format, NULL );
923 }
924 else
925 {
926 self->audio_format = self->video_format;
927 }
928 }
929 else if ( self->audio_index != -1 )
930 {
931 // We only have an audio context
932 self->audio_format = self->video_format;
933 self->video_format = NULL;
934 }
935 else if ( self->video_index == -1 )
936 {
937 // Something has gone wrong
938 error = -1;
939 }
940 if ( self->audio_format && !self->audio_streams )
941 get_audio_streams_info( self );
942
943 #ifdef AVFILTER
944 // Setup autorotate filters.
945 if (self->video_index != -1) {
946 self->autorotate = !mlt_properties_get(properties, "autorotate") || mlt_properties_get_int(properties, "autorotate");
947 if (!test_open && self->autorotate && !self->vfilter_graph) {
948 double theta = get_rotation(self->video_format->streams[self->video_index]);
949
950 if (fabs(theta - 90) < 1.0) {
951 error = ( setup_video_filters(self) < 0 );
952 AVFilterContext *last_filter = self->vfilter_out;
953 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "transpose", "clock") < 0 );
954 if (!error) error = ( avfilter_link(self->vfilter_in, 0, last_filter, 0) < 0 );
955 if (!error) error = ( avfilter_graph_config(self->vfilter_graph, NULL) < 0 );
956 } else if (fabs(theta - 180) < 1.0) {
957 error = ( setup_video_filters(self) < 0 );
958 AVFilterContext *last_filter = self->vfilter_out;
959 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "hflip", NULL) < 0 );
960 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "vflip", NULL) < 0 );
961 if (!error) error = ( avfilter_link(self->vfilter_in, 0, last_filter, 0) < 0 );
962 if (!error) error = ( avfilter_graph_config(self->vfilter_graph, NULL) < 0 );
963 } else if (fabs(theta - 270) < 1.0) {
964 error = ( setup_video_filters(self) < 0 );
965 AVFilterContext *last_filter = self->vfilter_out;
966 if (!error) error = ( insert_filter(self->vfilter_graph, &last_filter, "transpose", "cclock") < 0 );
967 if (!error) error = ( avfilter_link(self->vfilter_in, 0, last_filter, 0) < 0 );
968 if (!error) error = ( avfilter_graph_config(self->vfilter_graph, NULL) < 0 );
969 }
970 }
971 }
972 #endif
973 }
974 }
975 }
976 av_dict_free( ¶ms );
977 free( filename );
978 if ( !error )
979 {
980 self->apackets = mlt_deque_init();
981 self->vpackets = mlt_deque_init();
982 }
983
984 if ( self->dummy_context )
985 {
986 pthread_mutex_lock( &self->open_mutex );
987 avformat_close_input( &self->dummy_context );
988 self->dummy_context = NULL;
989 pthread_mutex_unlock( &self->open_mutex );
990 }
991
992 // Unlock the service
993 if ( take_lock )
994 {
995 pthread_mutex_unlock( &self->audio_mutex );
996 pthread_mutex_unlock( &self->video_mutex );
997 }
998 mlt_events_unblock( properties, self->parent );
999
1000 return error;
1001 }
1002
prepare_reopen(producer_avformat self)1003 static void prepare_reopen( producer_avformat self )
1004 {
1005 mlt_service_lock( MLT_PRODUCER_SERVICE( self->parent ) );
1006 pthread_mutex_lock( &self->audio_mutex );
1007 pthread_mutex_lock( &self->open_mutex );
1008
1009 int i;
1010 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
1011 {
1012 mlt_pool_release( self->audio_buffer[i] );
1013 self->audio_buffer[i] = NULL;
1014 av_free( self->decode_buffer[i] );
1015 self->decode_buffer[i] = NULL;
1016 if ( self->audio_codec[i] )
1017 avcodec_close( self->audio_codec[i] );
1018 self->audio_codec[i] = NULL;
1019 }
1020 if ( self->video_codec )
1021 avcodec_close( self->video_codec );
1022 self->video_codec = NULL;
1023 av_frame_unref( self->video_frame );
1024 #if USE_HWACCEL
1025 av_buffer_unref( &self->hwaccel.device_ctx );
1026 self->hwaccel.device_ctx = NULL;
1027 #endif
1028 if ( self->seekable && self->audio_format )
1029 avformat_close_input( &self->audio_format );
1030 if ( self->video_format )
1031 avformat_close_input( &self->video_format );
1032 self->audio_format = NULL;
1033 self->video_format = NULL;
1034 #ifdef AVFILTER
1035 avfilter_graph_free( &self->vfilter_graph );
1036 #endif
1037 pthread_mutex_unlock( &self->open_mutex );
1038
1039 // Cleanup the packet queues
1040 AVPacket *pkt;
1041 if ( self->apackets )
1042 {
1043 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
1044 {
1045 av_packet_unref( pkt );
1046 free( pkt );
1047 }
1048 mlt_deque_close( self->apackets );
1049 self->apackets = NULL;
1050 }
1051 if ( self->vpackets )
1052 {
1053 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
1054 {
1055 av_packet_unref( pkt );
1056 free( pkt );
1057 }
1058 mlt_deque_close( self->vpackets );
1059 self->vpackets = NULL;
1060 }
1061 pthread_mutex_unlock( &self->audio_mutex );
1062 mlt_service_unlock( MLT_PRODUCER_SERVICE( self->parent ) );
1063 }
1064
best_pts(producer_avformat self,int64_t pts,int64_t dts)1065 static int64_t best_pts( producer_avformat self, int64_t pts, int64_t dts )
1066 {
1067 self->invalid_pts_counter += pts == AV_NOPTS_VALUE;
1068 self->invalid_dts_counter += dts == AV_NOPTS_VALUE;
1069 if ( ( self->invalid_pts_counter <= self->invalid_dts_counter
1070 || dts == AV_NOPTS_VALUE ) && pts != AV_NOPTS_VALUE )
1071 return pts;
1072 else
1073 return dts;
1074 }
1075
find_first_pts(producer_avformat self,int video_index)1076 static void find_first_pts( producer_avformat self, int video_index )
1077 {
1078 // find initial PTS
1079 AVFormatContext *context = self->video_format? self->video_format : self->audio_format;
1080 int ret = 0;
1081 int pkt_countdown = 500; // check max 500 packets for first video keyframe PTS
1082 int vfr_countdown = 20; // check max 20 video frames for VFR
1083 int vfr_counter = 0; // counts the number of frame duration changes
1084 AVPacket pkt;
1085 int64_t prev_pkt_duration = AV_NOPTS_VALUE;
1086
1087 av_init_packet( &pkt );
1088 while ( ret >= 0 && pkt_countdown-- > 0 &&
1089 ( self->first_pts == AV_NOPTS_VALUE || ( vfr_counter < VFR_THRESHOLD && vfr_countdown > 0 ) ) )
1090 {
1091 ret = av_read_frame( context, &pkt );
1092 if ( ret >= 0 && pkt.stream_index == video_index )
1093 {
1094 // Variable frame rate check
1095 if ( pkt.duration != AV_NOPTS_VALUE && pkt.duration != prev_pkt_duration ) {
1096 mlt_log_verbose( MLT_PRODUCER_SERVICE(self->parent), "checking VFR: pkt.duration %"PRId64"\n", pkt.duration );
1097 if ( prev_pkt_duration != AV_NOPTS_VALUE )
1098 ++vfr_counter;
1099 }
1100 prev_pkt_duration = pkt.duration;
1101 vfr_countdown--;
1102
1103 // Finding PTS of first video key frame
1104 if ( ( pkt.flags & AV_PKT_FLAG_KEY ) && self->first_pts == AV_NOPTS_VALUE )
1105 {
1106 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
1107 "first_pts %"PRId64" dts %"PRId64" pts_dts_delta %d\n",
1108 pkt.pts, pkt.dts, (int)(pkt.pts - pkt.dts) );
1109 if ( pkt.dts != AV_NOPTS_VALUE && pkt.dts < 0 )
1110 // Decoding Time Stamps with negative values are reported by ffmpeg code for
1111 // (at least) MP4 files containing h.264 video using b-frames.
1112 // For reasons not understood yet, the first PTS computed then is that of the
1113 // third frame, causing MLT to display the third frame as if it was the first.
1114 // This if-clause is meant to catch and work around this issue - if there is
1115 // a valid, but negative DTS value, we just guess that the first valid
1116 // Presentation Time Stamp is == 0.
1117 self->first_pts = 0;
1118 else
1119 self->first_pts = best_pts( self, pkt.pts, pkt.dts );
1120 }
1121 }
1122 av_packet_unref( &pkt );
1123 }
1124 if ( vfr_counter >= VFR_THRESHOLD )
1125 mlt_properties_set_int( MLT_PRODUCER_PROPERTIES(self->parent), "meta.media.variable_frame_rate", 1 );
1126 av_seek_frame( context, -1, 0, AVSEEK_FLAG_BACKWARD );
1127 }
1128
seek_video(producer_avformat self,mlt_position position,int64_t req_position,int preseek)1129 static int seek_video( producer_avformat self, mlt_position position,
1130 int64_t req_position, int preseek )
1131 {
1132 mlt_producer producer = self->parent;
1133 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1134 int paused = 0;
1135 int seek_threshold = mlt_properties_get_int( properties, "seek_threshold" );
1136 if ( seek_threshold <= 0 ) seek_threshold = 12;
1137
1138 pthread_mutex_lock( &self->packets_mutex );
1139
1140 if ( self->video_seekable && ( position != self->video_expected || self->last_position < 0 ) )
1141 {
1142
1143 // Fetch the video format context
1144 AVFormatContext *context = self->video_format;
1145
1146 // Get the video stream
1147 AVStream *stream = context->streams[ self->video_index ];
1148
1149 // Get codec context
1150 AVCodecContext *codec_context = stream->codec;
1151
1152 // We may want to use the source fps if available
1153 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1154 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1155
1156 if ( self->first_pts == AV_NOPTS_VALUE && self->last_position == POSITION_INITIAL )
1157 find_first_pts( self, self->video_index );
1158
1159 if ( self->video_frame && position + 1 == self->video_expected )
1160 {
1161 // We're paused - use last image
1162 paused = 1;
1163 }
1164 else if ( position < self->video_expected || position - self->video_expected >= seek_threshold || self->last_position < 0 )
1165 {
1166 // Calculate the timestamp for the requested frame
1167 int64_t timestamp = req_position / ( av_q2d( self->video_time_base ) * source_fps );
1168 if ( req_position <= 0 )
1169 timestamp = 0;
1170 else if ( self->first_pts != AV_NOPTS_VALUE )
1171 timestamp += self->first_pts;
1172 else if ( context->start_time != AV_NOPTS_VALUE )
1173 timestamp += context->start_time;
1174 if ( preseek && av_q2d( self->video_time_base ) != 0 )
1175 timestamp -= 2 / av_q2d( self->video_time_base );
1176 if ( timestamp < 0 )
1177 timestamp = 0;
1178 mlt_log_debug( MLT_PRODUCER_SERVICE(producer), "seeking timestamp %"PRId64" position " MLT_POSITION_FMT " expected "MLT_POSITION_FMT" last_pos %"PRId64"\n",
1179 timestamp, position, self->video_expected, self->last_position );
1180
1181 // Seek to the timestamp
1182 codec_context->skip_loop_filter = AVDISCARD_NONREF;
1183 av_seek_frame( context, self->video_index, timestamp, AVSEEK_FLAG_BACKWARD );
1184
1185 // flush any pictures still in decode buffer
1186 avcodec_flush_buffers( codec_context );
1187 self->video_send_result = 0;
1188
1189 // Remove the cached info relating to the previous position
1190 self->current_position = POSITION_INVALID;
1191 self->last_position = POSITION_INVALID;
1192 av_frame_unref(self->video_frame);
1193 }
1194 }
1195 pthread_mutex_unlock( &self->packets_mutex );
1196 return paused;
1197 }
1198
1199 /** Convert a frame position to a time code.
1200 */
1201
producer_time_of_frame(mlt_producer producer,mlt_position position)1202 static double producer_time_of_frame( mlt_producer producer, mlt_position position )
1203 {
1204 return ( double )position / mlt_producer_get_fps( producer );
1205 }
1206
1207 // Collect information about all audio streams
1208
get_audio_streams_info(producer_avformat self)1209 static void get_audio_streams_info( producer_avformat self )
1210 {
1211 // Fetch the audio format context
1212 AVFormatContext *context = self->audio_format;
1213 unsigned int i;
1214
1215 for ( i = 0;
1216 i < context->nb_streams;
1217 i++ )
1218 {
1219 if ( context->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO )
1220 {
1221 AVCodecParameters *codec_params = context->streams[i]->codecpar;
1222 AVCodec *codec = avcodec_find_decoder( codec_params->codec_id );
1223
1224 // If we don't have a codec and we can't initialise it, we can't do much more...
1225 pthread_mutex_lock( &self->open_mutex );
1226 if ( codec && avcodec_open2( context->streams[i]->codec, codec, NULL ) >= 0 )
1227 {
1228 self->audio_streams++;
1229 self->audio_max_stream = i;
1230 self->total_channels += codec_params->channels;
1231 if ( codec_params->channels > self->max_channel )
1232 self->max_channel = codec_params->channels;
1233 if ( codec_params->sample_rate > self->max_frequency )
1234 self->max_frequency = codec_params->sample_rate;
1235 avcodec_close( context->streams[i]->codec );
1236 }
1237 pthread_mutex_unlock( &self->open_mutex );
1238 }
1239 }
1240 mlt_log_verbose( NULL, "[producer avformat] audio: total_streams %d max_stream %d total_channels %d max_channels %d\n",
1241 self->audio_streams, self->audio_max_stream, self->total_channels, self->max_channel );
1242 }
1243
pick_image_format(enum AVPixelFormat pix_fmt)1244 static mlt_image_format pick_image_format( enum AVPixelFormat pix_fmt )
1245 {
1246 switch ( pix_fmt )
1247 {
1248 case AV_PIX_FMT_ARGB:
1249 case AV_PIX_FMT_RGBA:
1250 case AV_PIX_FMT_ABGR:
1251 case AV_PIX_FMT_BGRA:
1252 return mlt_image_rgb24a;
1253 case AV_PIX_FMT_YUV420P:
1254 case AV_PIX_FMT_YUVJ420P:
1255 case AV_PIX_FMT_YUVA420P:
1256 return mlt_image_yuv420p;
1257 case AV_PIX_FMT_RGB24:
1258 case AV_PIX_FMT_BGR24:
1259 case AV_PIX_FMT_GRAY8:
1260 case AV_PIX_FMT_MONOWHITE:
1261 case AV_PIX_FMT_MONOBLACK:
1262 case AV_PIX_FMT_RGB8:
1263 case AV_PIX_FMT_BGR8:
1264 #if defined(FFUDIV)
1265 case AV_PIX_FMT_BAYER_RGGB16LE:
1266 return mlt_image_rgb24;
1267 #endif
1268 default:
1269 return mlt_image_yuv422;
1270 }
1271 }
1272
pick_audio_format(int sample_fmt)1273 static mlt_audio_format pick_audio_format( int sample_fmt )
1274 {
1275 switch ( sample_fmt )
1276 {
1277 // interleaved
1278 case AV_SAMPLE_FMT_U8:
1279 return mlt_audio_u8;
1280 case AV_SAMPLE_FMT_S16:
1281 return mlt_audio_s16;
1282 case AV_SAMPLE_FMT_S32:
1283 return mlt_audio_s32le;
1284 case AV_SAMPLE_FMT_FLT:
1285 return mlt_audio_f32le;
1286 // planar - this producer converts planar to interleaved
1287 case AV_SAMPLE_FMT_U8P:
1288 return mlt_audio_u8;
1289 case AV_SAMPLE_FMT_S16P:
1290 return mlt_audio_s16;
1291 case AV_SAMPLE_FMT_S32P:
1292 return mlt_audio_s32le;
1293 case AV_SAMPLE_FMT_FLTP:
1294 return mlt_audio_f32le;
1295 default:
1296 return mlt_audio_none;
1297 }
1298 }
1299
1300 /**
1301 * Handle deprecated pixel format (JPEG range in YUV420P for example).
1302 *
1303 * Replace pix_fmt with the official pixel format to use.
1304 * @return 0 if no pix_fmt replacement, 1 otherwise
1305 */
pick_av_pixel_format(int * pix_fmt)1306 static int pick_av_pixel_format( int *pix_fmt )
1307 {
1308 #if defined(FFUDIV)
1309 switch (*pix_fmt)
1310 {
1311 case AV_PIX_FMT_YUVJ420P:
1312 *pix_fmt = AV_PIX_FMT_YUV420P;
1313 return 1;
1314 case AV_PIX_FMT_YUVJ411P:
1315 *pix_fmt = AV_PIX_FMT_YUV411P;
1316 return 1;
1317 case AV_PIX_FMT_YUVJ422P:
1318 *pix_fmt = AV_PIX_FMT_YUV422P;
1319 return 1;
1320 case AV_PIX_FMT_YUVJ444P:
1321 *pix_fmt = AV_PIX_FMT_YUV444P;
1322 return 1;
1323 case AV_PIX_FMT_YUVJ440P:
1324 *pix_fmt = AV_PIX_FMT_YUV440P;
1325 return 1;
1326 }
1327 #endif
1328 return 0;
1329 }
1330
1331 #if defined(FFUDIV) && (LIBSWSCALE_VERSION_INT >= ((3<<16)+(1<<8)+101))
1332 struct sliced_pix_fmt_conv_t
1333 {
1334 int width, height, slice_w;
1335 AVFrame *frame;
1336 uint8_t *out_data[4];
1337 int out_stride[4];
1338 enum AVPixelFormat src_format, dst_format;
1339 const AVPixFmtDescriptor *src_desc, *dst_desc;
1340 int flags, src_colorspace, dst_colorspace, src_full_range, dst_full_range;
1341 };
1342
sliced_h_pix_fmt_conv_proc(int id,int idx,int jobs,void * cookie)1343 static int sliced_h_pix_fmt_conv_proc( int id, int idx, int jobs, void* cookie )
1344 {
1345 uint8_t *out[4];
1346 const uint8_t *in[4];
1347 int in_stride[4], out_stride[4];
1348 int src_v_chr_pos = -513, dst_v_chr_pos = -513, ret, i, slice_x, slice_w, h, mul, field, slices, interlaced = 0;
1349
1350 struct SwsContext *sws;
1351 struct sliced_pix_fmt_conv_t* ctx = ( struct sliced_pix_fmt_conv_t* )cookie;
1352
1353 interlaced = ctx->frame->interlaced_frame;
1354 field = ( interlaced ) ? ( idx & 1 ) : 0;
1355 idx = ( interlaced ) ? ( idx / 2 ) : idx;
1356 slices = ( interlaced ) ? ( jobs / 2 ) : jobs;
1357 mul = ( interlaced ) ? 2 : 1;
1358 h = ctx->height >> !!interlaced;
1359 slice_w = ctx->slice_w;
1360 slice_x = slice_w * idx;
1361 slice_w = FFMIN( slice_w, ctx->width - slice_x );
1362
1363 if ( AV_PIX_FMT_YUV420P == ctx->src_format )
1364 src_v_chr_pos = ( !interlaced ) ? 128 : ( !field ) ? 64 : 192;
1365
1366 if ( AV_PIX_FMT_YUV420P == ctx->dst_format )
1367 dst_v_chr_pos = ( !interlaced ) ? 128 : ( !field ) ? 64 : 192;
1368
1369 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",
1370 __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 );
1371
1372 if ( slice_w <= 0 )
1373 return 0;
1374
1375 sws = sws_alloc_context();
1376
1377 av_opt_set_int( sws, "srcw", slice_w, 0 );
1378 av_opt_set_int( sws, "srch", h, 0 );
1379 av_opt_set_int( sws, "src_format", ctx->src_format, 0 );
1380 av_opt_set_int( sws, "dstw", slice_w, 0 );
1381 av_opt_set_int( sws, "dsth", h, 0 );
1382 av_opt_set_int( sws, "dst_format", ctx->dst_format, 0 );
1383 av_opt_set_int( sws, "sws_flags", ctx->flags, 0 );
1384
1385 av_opt_set_int( sws, "src_h_chr_pos", -513, 0 );
1386 av_opt_set_int( sws, "src_v_chr_pos", src_v_chr_pos, 0 );
1387 av_opt_set_int( sws, "dst_h_chr_pos", -513, 0 );
1388 av_opt_set_int( sws, "dst_v_chr_pos", dst_v_chr_pos, 0 );
1389
1390 if ( ( ret = sws_init_context( sws, NULL, NULL ) ) < 0 )
1391 {
1392 mlt_log_error( NULL, "%s:%d: sws_init_context failed, ret=%d\n", __FUNCTION__, __LINE__, ret );
1393 sws_freeContext( sws );
1394 return 0;
1395 }
1396
1397 mlt_set_luma_transfer( sws, ctx->src_colorspace, ctx->dst_colorspace, ctx->src_full_range, ctx->dst_full_range );
1398
1399 #if LIBAVUTIL_VERSION_INT < AV_VERSION_INT(55, 0, 100)
1400 #define PIX_DESC_BPP(DESC) (DESC.step_minus1 + 1)
1401 #else
1402 #define PIX_DESC_BPP(DESC) (DESC.step)
1403 #endif
1404
1405 for( i = 0; i < 4; i++ )
1406 {
1407 int in_offset = (AV_PIX_FMT_FLAG_PLANAR & ctx->src_desc->flags)
1408 ? ( ( 1 == i || 2 == i ) ? ( slice_x >> ctx->src_desc->log2_chroma_w ) : slice_x )
1409 : ( ( 0 == i ) ? slice_x : 0 );
1410
1411 int out_offset = (AV_PIX_FMT_FLAG_PLANAR & ctx->dst_desc->flags)
1412 ? ( ( 1 == i || 2 == i ) ? ( slice_x >> ctx->dst_desc->log2_chroma_w ) : slice_x )
1413 : ( ( 0 == i ) ? slice_x : 0 );
1414
1415 in_offset *= PIX_DESC_BPP(ctx->src_desc->comp[i]);
1416 out_offset *= PIX_DESC_BPP(ctx->dst_desc->comp[i]);
1417
1418 in_stride[i] = ctx->frame->linesize[i] * mul;
1419 out_stride[i] = ctx->out_stride[i] * mul;
1420
1421 in[i] = ctx->frame->data[i] + ctx->frame->linesize[i] * field + in_offset;
1422 out[i] = ctx->out_data[i] + ctx->out_stride[i] * field + out_offset;
1423 }
1424
1425 sws_scale( sws, in, in_stride, 0, h, out, out_stride );
1426
1427 sws_freeContext( sws );
1428
1429 return 0;
1430 }
1431 #endif
1432
1433 // 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)1434 static int convert_image( producer_avformat self, AVFrame *frame, uint8_t *buffer, int pix_fmt,
1435 mlt_image_format *format, int width, int height, uint8_t **alpha )
1436 {
1437 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
1438 int result = self->yuv_colorspace;
1439
1440 mlt_log_timings_begin();
1441
1442 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "%s @ %dx%d space %d->%d\n",
1443 mlt_image_format_name( *format ),
1444 width, height, self->yuv_colorspace, profile->colorspace );
1445
1446 // extract alpha from planar formats
1447 if ( ( pix_fmt == AV_PIX_FMT_YUVA420P
1448 #if defined(FFUDIV)
1449 || pix_fmt == AV_PIX_FMT_YUVA444P
1450 #endif
1451 ) &&
1452 *format != mlt_image_rgb24a && *format != mlt_image_opengl &&
1453 frame->data[3] && frame->linesize[3] )
1454 {
1455 int i;
1456 uint8_t *src, *dst;
1457
1458 dst = *alpha = mlt_pool_alloc( width * height );
1459 src = frame->data[3];
1460
1461 for ( i = 0; i < height; dst += width, src += frame->linesize[3], i++ )
1462 memcpy( dst, src, FFMIN( width, frame->linesize[3] ) );
1463 }
1464
1465 int src_pix_fmt = pix_fmt;
1466 pick_av_pixel_format( &src_pix_fmt );
1467 if ( *format == mlt_image_yuv420p )
1468 {
1469 // This is a special case. Movit wants the full range, if available.
1470 // Thankfully, there is not much other use of yuv420p except consumer
1471 // avformat with no filters and explicitly requested.
1472 #if defined(FFUDIV)
1473 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_YUV420P);
1474 struct SwsContext *context = sws_getContext(width, height, src_pix_fmt,
1475 width, height, AV_PIX_FMT_YUV420P, flags, NULL, NULL, NULL);
1476 #else
1477 int dst_pix_fmt = self->full_luma ? AV_PIX_FMT_YUVJ420P : AV_PIX_FMT_YUV420P;
1478 int flags = mlt_get_sws_flags(width, height, pix_fmt, width, height, dst_pix_fmt);
1479 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1480 width, height, dst_pix_fmt,
1481 flags, NULL, NULL, NULL);
1482 #endif
1483
1484 uint8_t *out_data[4];
1485 int out_stride[4];
1486 out_data[0] = buffer;
1487 out_data[1] = buffer + width * height;
1488 out_data[2] = buffer + ( 5 * width * height ) / 4;
1489 out_stride[0] = width;
1490 out_stride[1] = width >> 1;
1491 out_stride[2] = width >> 1;
1492 if ( !mlt_set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma, self->full_luma ) )
1493 result = profile->colorspace;
1494 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1495 out_data, out_stride);
1496 sws_freeContext( context );
1497 }
1498 else if ( *format == mlt_image_rgb24 )
1499 {
1500 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_RGB24);
1501 struct SwsContext *context = sws_getContext( width, height, src_pix_fmt,
1502 width, height, AV_PIX_FMT_RGB24, flags, NULL, NULL, NULL);
1503 uint8_t *out_data[4];
1504 int out_stride[4];
1505 av_image_fill_arrays(out_data, out_stride, buffer, AV_PIX_FMT_RGB24, width, height, IMAGE_ALIGN);
1506 // libswscale wants the RGB colorspace to be SWS_CS_DEFAULT, which is = SWS_CS_ITU601.
1507 mlt_set_luma_transfer( context, self->yuv_colorspace, 601, self->full_luma, 0 );
1508 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1509 out_data, out_stride);
1510 sws_freeContext( context );
1511 }
1512 else if ( *format == mlt_image_rgb24a || *format == mlt_image_opengl )
1513 {
1514 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_RGBA);
1515 struct SwsContext *context = sws_getContext( width, height, src_pix_fmt,
1516 width, height, AV_PIX_FMT_RGBA, flags, NULL, NULL, NULL);
1517 uint8_t *out_data[4];
1518 int out_stride[4];
1519 av_image_fill_arrays(out_data, out_stride, buffer, AV_PIX_FMT_RGBA, width, height, IMAGE_ALIGN);
1520 // libswscale wants the RGB colorspace to be SWS_CS_DEFAULT, which is = SWS_CS_ITU601.
1521 mlt_set_luma_transfer( context, self->yuv_colorspace, 601, self->full_luma, 0 );
1522 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1523 out_data, out_stride);
1524 sws_freeContext( context );
1525 }
1526 else
1527 #if defined(FFUDIV) && (LIBSWSCALE_VERSION_INT >= ((3<<16)+(1<<8)+101))
1528 {
1529 int i, c;
1530 struct sliced_pix_fmt_conv_t ctx =
1531 {
1532 .width = width,
1533 .height = height,
1534 .frame = frame,
1535 .dst_format = AV_PIX_FMT_YUYV422,
1536 .src_colorspace = self->yuv_colorspace,
1537 .dst_colorspace = profile->colorspace,
1538 .src_full_range = self->full_luma,
1539 .dst_full_range = 0,
1540 };
1541 ctx.src_format = (self->full_luma && src_pix_fmt == AV_PIX_FMT_YUV422P) ? AV_PIX_FMT_YUVJ422P : src_pix_fmt;
1542 ctx.src_desc = av_pix_fmt_desc_get( ctx.src_format );
1543 ctx.dst_desc = av_pix_fmt_desc_get( ctx.dst_format );
1544 ctx.flags = mlt_get_sws_flags(width, height, ctx.src_format, width, height, ctx.dst_format);
1545
1546 av_image_fill_arrays(ctx.out_data, ctx.out_stride, buffer, ctx.dst_format, width, height, IMAGE_ALIGN);
1547
1548 int sliced = !getenv("MLT_AVFORMAT_SLICED_PIXFMT_DISABLE");
1549 if ( sliced ) {
1550 ctx.slice_w = ( width < 1000 )
1551 ? ( 256 >> frame->interlaced_frame )
1552 : ( 512 >> frame->interlaced_frame );
1553 } else {
1554 ctx.slice_w = width;
1555 }
1556
1557 c = ( width + ctx.slice_w - 1 ) / ctx.slice_w;
1558 int last_slice_w = width - ctx.slice_w * (c - 1);
1559
1560 if ( sliced && (last_slice_w % 8) == 0 && !(ctx.src_format == AV_PIX_FMT_YUV422P && last_slice_w % 16) ) {
1561 c *= frame->interlaced_frame ? 2 : 1;
1562 mlt_slices_run_normal( c, sliced_h_pix_fmt_conv_proc, &ctx );
1563 } else {
1564 c = frame->interlaced_frame ? 2 : 1;
1565 ctx.slice_w = width;
1566 for ( i = 0 ; i < c; i++ )
1567 sliced_h_pix_fmt_conv_proc( i, i, c, &ctx );
1568 }
1569
1570 result = profile->colorspace;
1571 }
1572 #else
1573 {
1574 #if defined(FFUDIV)
1575 int flags = mlt_get_sws_flags(width, height, src_pix_fmt, width, height, AV_PIX_FMT_YUYV422);
1576 struct SwsContext *context = sws_getContext( width, height, src_pix_fmt,
1577 width, height, AV_PIX_FMT_YUYV422, flags, NULL, NULL, NULL);
1578 #else
1579 int flags = mlt_get_sws_flags(width, height, pix_fmt, width, height, AV_PIX_FMT_YUYV422);
1580 struct SwsContext *context = sws_getContext( width, height, pix_fmt,
1581 width, height, AV_PIX_FMT_YUYV422, flags, NULL, NULL, NULL);
1582 #endif
1583 AVPicture output;
1584 avpicture_fill( &output, buffer, AV_PIX_FMT_YUYV422, width, height );
1585 if ( !mlt_set_luma_transfer( context, self->yuv_colorspace, profile->colorspace, self->full_luma, 0 ) )
1586 result = profile->colorspace;
1587 sws_scale( context, (const uint8_t* const*) frame->data, frame->linesize, 0, height,
1588 output.data, output.linesize);
1589 sws_freeContext( context );
1590 }
1591 #endif
1592 mlt_log_timings_end( NULL, __FUNCTION__ );
1593
1594 return result;
1595 }
1596
set_image_size(producer_avformat self,int * width,int * height)1597 static void set_image_size( producer_avformat self, int *width, int *height )
1598 {
1599 double dar = mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE(self->parent) ) );
1600 double theta = self->autorotate? get_rotation( self->video_format->streams[self->video_index] ) : 0.0;
1601 if ( fabs(theta - 90.0) < 1.0 || fabs(theta - 270.0) < 1.0 )
1602 {
1603 *height = self->video_codec->width;
1604 // Workaround 1088 encodings missing cropping info.
1605 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 )
1606 *width = 1080;
1607 else
1608 *width = self->video_codec->height;
1609 } else {
1610 *width = self->video_codec->width;
1611 // Workaround 1088 encodings missing cropping info.
1612 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 )
1613 *height = 1080;
1614 else
1615 *height = self->video_codec->height;
1616 }
1617 }
1618
1619 /** Allocate the image buffer and set it on the frame.
1620 */
1621
allocate_buffer(mlt_frame frame,AVCodecParameters * codec_params,uint8_t ** buffer,mlt_image_format format,int width,int height)1622 static int allocate_buffer( mlt_frame frame, AVCodecParameters *codec_params, uint8_t **buffer, mlt_image_format format, int width, int height )
1623 {
1624 int size = 0;
1625
1626 if ( codec_params->width == 0 || codec_params->height == 0 )
1627 return size;
1628
1629 size = mlt_image_format_size( format, width, height, NULL );
1630 *buffer = mlt_pool_alloc( size );
1631 if ( *buffer )
1632 mlt_frame_set_image( frame, *buffer, size, mlt_pool_release );
1633 else
1634 size = 0;
1635
1636 return size;
1637 }
1638
ignore_send_packet_result(int result)1639 static int ignore_send_packet_result(int result)
1640 {
1641 return result >= 0 || result == AVERROR(EAGAIN) || result == AVERROR_EOF || result == AVERROR_INVALIDDATA;
1642 }
1643
1644 /** Get an image from a frame.
1645 */
1646
producer_get_image(mlt_frame frame,uint8_t ** buffer,mlt_image_format * format,int * width,int * height,int writable)1647 static int producer_get_image( mlt_frame frame, uint8_t **buffer, mlt_image_format *format, int *width, int *height, int writable )
1648 {
1649 // Get the producer
1650 (void) writable; // unused
1651 producer_avformat self = mlt_frame_pop_service( frame );
1652 mlt_producer producer = self->parent;
1653
1654 // Get the properties from the frame
1655 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
1656
1657 // Obtain the frame number of this frame
1658 mlt_position position = mlt_frame_original_position( frame );
1659
1660 // Get the producer properties
1661 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
1662
1663 uint8_t *alpha = NULL;
1664 int got_picture = 0;
1665 int image_size = 0;
1666
1667 pthread_mutex_lock( &self->video_mutex );
1668 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1669 mlt_log_timings_begin();
1670
1671 // Fetch the video format context
1672 AVFormatContext *context = self->video_format;
1673 AVCodecParameters *codec_params = NULL;
1674 if ( !context )
1675 goto exit_get_image;
1676
1677 // Get the video stream
1678 AVStream *stream = context->streams[ self->video_index ];
1679
1680 // Get codec context
1681 AVCodecContext *codec_context = stream->codec;
1682 codec_params = stream->codecpar;
1683
1684 // Always use the image cache for album art.
1685 int is_album_art = ((codec_context->codec_id == AV_CODEC_ID_MJPEG
1686 || codec_context->codec_id == AV_CODEC_ID_GIF
1687 || codec_context->codec_id == AV_CODEC_ID_PNG)
1688 && mlt_properties_get_int(properties, "meta.media.frame_rate_num") == 90000
1689 && mlt_properties_get_int(properties, "meta.media.frame_rate_den") == 1);
1690 if (is_album_art)
1691 position = 0;
1692
1693 // Get the image cache
1694 if ( ! self->image_cache )
1695 {
1696 // if cache size supplied by environment variable
1697 int cache_supplied = getenv( "MLT_AVFORMAT_CACHE" ) != NULL;
1698 int cache_size = cache_supplied? atoi( getenv( "MLT_AVFORMAT_CACHE" ) ) : 0;
1699
1700 // cache size supplied via property
1701 if ( mlt_properties_get( properties, "cache" ) )
1702 {
1703 cache_supplied = 1;
1704 cache_size = mlt_properties_get_int( properties, "cache" );
1705 }
1706 if ( mlt_properties_get_int( properties, "noimagecache" ) )
1707 {
1708 cache_supplied = 1;
1709 cache_size = 0;
1710 }
1711 // create cache if not disabled
1712 if ( !cache_supplied || cache_size > 0 )
1713 self->image_cache = mlt_cache_init();
1714 // set cache size if supplied
1715 if ( self->image_cache && cache_supplied )
1716 mlt_cache_set_size( self->image_cache, cache_size );
1717 }
1718 if ( self->image_cache )
1719 {
1720 mlt_frame original = mlt_cache_get_frame( self->image_cache, position );
1721 if ( original )
1722 {
1723 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
1724 int size = 0;
1725
1726 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
1727 if (*buffer)
1728 mlt_frame_set_alpha( frame, *buffer, size, NULL );
1729 *buffer = mlt_properties_get_data( orig_props, "image", &size );
1730 mlt_frame_set_image( frame, *buffer, size, NULL );
1731 mlt_properties_set_data( frame_properties, "avformat.image_cache", original, 0, (mlt_destructor) mlt_frame_close, NULL );
1732 *format = mlt_properties_get_int( orig_props, "format" );
1733 set_image_size( self, width, height );
1734 mlt_properties_pass_property(frame_properties, orig_props, "colorspace");
1735 got_picture = 1;
1736 goto exit_get_image;
1737 }
1738 }
1739 // Cache miss
1740
1741 // We may want to use the source fps if available
1742 double source_fps = mlt_properties_get_double( properties, "meta.media.frame_rate_num" ) /
1743 mlt_properties_get_double( properties, "meta.media.frame_rate_den" );
1744
1745 // This is the physical frame position in the source
1746 int64_t req_position = ( int64_t )( position / mlt_producer_get_fps( producer ) * source_fps + 0.5 );
1747
1748 // Determines if we have to decode all frames in a sequence - when there temporal compression is used.
1749 const AVCodecDescriptor *descriptor = avcodec_descriptor_get( codec_params->codec_id );
1750 int must_decode = descriptor && !( descriptor->props & AV_CODEC_PROP_INTRA_ONLY );
1751
1752 double delay = mlt_properties_get_double( properties, "video_delay" );
1753
1754 // Seek if necessary
1755 double speed = mlt_producer_get_speed(producer);
1756 int preseek = must_decode && codec_context->has_b_frames && speed >= 0.0 && speed <= 1.0;
1757 int paused = seek_video( self, position, req_position, preseek );
1758
1759 // Seek might have reopened the file
1760 context = self->video_format;
1761 stream = context->streams[ self->video_index ];
1762 codec_context = stream->codec;
1763 codec_params = stream->codecpar;
1764 if ( *format == mlt_image_none || *format == mlt_image_glsl ||
1765 codec_params->format == AV_PIX_FMT_ARGB ||
1766 codec_params->format == AV_PIX_FMT_RGBA ||
1767 codec_params->format == AV_PIX_FMT_ABGR ||
1768 codec_params->format == AV_PIX_FMT_BGRA )
1769 *format = pick_image_format( codec_params->format );
1770 #if defined(FFUDIV)
1771 else if ( codec_params->format == AV_PIX_FMT_BAYER_RGGB16LE ) {
1772 if ( *format == mlt_image_yuv422 )
1773 *format = mlt_image_yuv420p;
1774 else if ( *format == mlt_image_rgb24a )
1775 *format = mlt_image_rgb24;
1776 }
1777 #endif
1778 else if ( codec_params->format == AV_PIX_FMT_YUVA444P10LE
1779 #if LIBAVUTIL_VERSION_INT >= AV_VERSION_INT(56,0,0)
1780 || codec_params->format == AV_PIX_FMT_GBRAP10LE
1781 || codec_params->format == AV_PIX_FMT_GBRAP12LE
1782 #endif
1783 )
1784 *format = mlt_image_rgb24a;
1785
1786 // Duplicate the last image if necessary
1787 if ( self->video_frame && self->video_frame->linesize[0]
1788 && (self->pkt.stream_index == self->video_index )
1789 && ( paused || self->current_position >= req_position ) )
1790 {
1791 // Duplicate it
1792 set_image_size( self, width, height );
1793 if ( ( image_size = allocate_buffer( frame, codec_params, buffer, *format, *width, *height ) ) )
1794 {
1795 int yuv_colorspace;
1796 #if USE_HWACCEL
1797 yuv_colorspace = convert_image( self, self->video_frame, *buffer, self->video_frame->format,
1798 format, *width, *height, &alpha );
1799 #else
1800 yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_params->format,
1801 format, *width, *height, &alpha );
1802 #endif
1803 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
1804 got_picture = 1;
1805 }
1806 }
1807 else
1808 {
1809 int64_t int_position = 0;
1810 int decode_errors = 0;
1811
1812 // Construct an AVFrame for YUV422 conversion
1813 if ( !self->video_frame )
1814 self->video_frame = av_frame_alloc();
1815 else
1816 av_frame_unref( self->video_frame );
1817
1818 while (!got_picture && ignore_send_packet_result(self->video_send_result))
1819 {
1820 if ( self->video_send_result != AVERROR( EAGAIN ) )
1821 {
1822 // Read a packet
1823 if ( self->pkt.stream_index == self->video_index )
1824 av_packet_unref( &self->pkt );
1825 av_init_packet( &self->pkt );
1826 pthread_mutex_lock( &self->packets_mutex );
1827 if ( mlt_deque_count( self->vpackets ) )
1828 {
1829 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->vpackets );
1830 av_packet_ref( &self->pkt, tmp );
1831 av_packet_free( &tmp );
1832 }
1833 else
1834 {
1835 int ret = av_read_frame( context, &self->pkt );
1836 if ( ret >= 0 && !self->video_seekable && self->pkt.stream_index == self->audio_index )
1837 {
1838 mlt_deque_push_back( self->apackets, av_packet_clone( &self->pkt ) );
1839 }
1840 else if ( ret < 0 )
1841 {
1842 if ( ret == AVERROR_EOF )
1843 {
1844 self->pkt.stream_index = self->video_index;
1845 } else
1846 {
1847 mlt_log_verbose( MLT_PRODUCER_SERVICE( producer ), "av_read_frame returned error %d inside get_image\n", ret );
1848 }
1849 if ( !self->video_seekable && mlt_properties_get_int( properties, "reconnect" ) )
1850 {
1851 // Try to reconnect to live sources by closing context and codecs,
1852 // and letting next call to get_frame() reopen.
1853 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
1854 prepare_reopen( self );
1855 mlt_service_lock( MLT_PRODUCER_SERVICE( producer ) );
1856 pthread_mutex_unlock( &self->packets_mutex );
1857 goto exit_get_image;
1858 }
1859 if ( !self->video_seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) )
1860 {
1861 mlt_log_fatal( MLT_PRODUCER_SERVICE( producer ), "Exiting with error due to disconnected source.\n" );
1862 exit( EXIT_FAILURE );
1863 }
1864 // Send null packets to drain decoder.
1865 self->pkt.size = 0;
1866 self->pkt.data = NULL;
1867 }
1868 }
1869 pthread_mutex_unlock( &self->packets_mutex );
1870 }
1871
1872 // We only deal with video from the selected video_index
1873 if ( self->pkt.stream_index == self->video_index )
1874 {
1875 int64_t pts = best_pts( self, self->pkt.pts, self->pkt.dts );
1876 if ( pts != AV_NOPTS_VALUE )
1877 {
1878 if ( !self->video_seekable && self->first_pts == AV_NOPTS_VALUE )
1879 self->first_pts = pts;
1880 if ( self->first_pts != AV_NOPTS_VALUE )
1881 pts -= self->first_pts;
1882 else if ( context->start_time != AV_NOPTS_VALUE )
1883 pts -= context->start_time;
1884 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1885 if ( int_position == self->last_position )
1886 int_position = self->last_position + 1;
1887 }
1888 mlt_log_debug( MLT_PRODUCER_SERVICE(producer),
1889 "V pkt.pts %"PRId64" pkt.dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
1890 self->pkt.pts, self->pkt.dts, req_position, self->current_position, int_position );
1891
1892 // Make a dumb assumption on streams that contain wild timestamps
1893 if ( llabs( req_position - int_position ) > 999 )
1894 {
1895 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), " WILD TIMESTAMP: "
1896 "pkt.pts=[%"PRId64"], pkt.dts=[%"PRId64"], req_position=[%"PRId64"], "
1897 "current_position=[%"PRId64"], int_position=[%"PRId64"], pts=[%"PRId64"] \n",
1898 self->pkt.pts, self->pkt.dts, req_position,
1899 self->current_position, int_position, pts );
1900 int_position = req_position;
1901 }
1902 self->last_position = int_position;
1903
1904 // Decode the image
1905 if ( must_decode || int_position >= req_position || !self->pkt.data )
1906 {
1907 codec_context->reordered_opaque = int_position;
1908 if ( int_position >= req_position )
1909 codec_context->skip_loop_filter = AVDISCARD_NONE;
1910 self->video_send_result = avcodec_send_packet( codec_context, &self->pkt );
1911 mlt_log_debug( MLT_PRODUCER_SERVICE( producer ), "decoded video packet with size %d => %d\n", self->pkt.size, self->video_send_result );
1912 // Note: decode may fail at the beginning of MPEGfile (B-frames referencing before first I-frame), so allow a few errors.
1913 if (!ignore_send_packet_result(self->video_send_result))
1914 {
1915 mlt_log_warning( MLT_PRODUCER_SERVICE( producer ), "video avcodec_send_packet failed with %d\n", self->video_send_result );
1916 }
1917 else
1918 {
1919 int error = avcodec_receive_frame( codec_context, self->video_frame );
1920 if ( error < 0 )
1921 {
1922 if ( error != AVERROR( EAGAIN ) && ++decode_errors > 10 )
1923 {
1924 mlt_log_warning( MLT_PRODUCER_SERVICE( producer ), "video decoding error %d\n", error );
1925 self->last_good_position = POSITION_INVALID;
1926 }
1927 }
1928 else
1929 {
1930 #if USE_HWACCEL
1931 if (self->hwaccel.device_ctx && self->video_frame->format == self->hwaccel.pix_fmt)
1932 {
1933 AVFrame *sw_video_frame = av_frame_alloc();
1934 int transfer_data_result = av_hwframe_transfer_data(sw_video_frame, self->video_frame, 0);
1935 if(transfer_data_result < 0)
1936 {
1937 mlt_log_error( MLT_PRODUCER_SERVICE(producer), "av_hwframe_transfer_data() failed %d\n", transfer_data_result);
1938 av_frame_free(&sw_video_frame);
1939 goto exit_get_image;
1940 }
1941 av_frame_copy_props(sw_video_frame, self->video_frame);
1942 sw_video_frame->width = self->video_frame->width;
1943 sw_video_frame->height = self->video_frame->height;
1944
1945 av_frame_unref(self->video_frame);
1946 av_frame_move_ref(self->video_frame, sw_video_frame);
1947 av_frame_free(&sw_video_frame);
1948 }
1949 #endif
1950 got_picture = 1;
1951 decode_errors = 0;
1952 }
1953 }
1954 }
1955
1956 if ( got_picture )
1957 {
1958 // Get position of reordered frame
1959 int_position = self->video_frame->reordered_opaque;
1960 pts = best_pts( self, self->video_frame->pts, self->video_frame->pkt_dts );
1961 if ( pts != AV_NOPTS_VALUE )
1962 {
1963 // Some streams are not marking their key frames even though
1964 // there are I frames, and find_first_pts() fails as a result.
1965 // Try to set first_pts here after getting pict_type.
1966 if ( self->first_pts == AV_NOPTS_VALUE &&
1967 (self->video_frame->key_frame || self->video_frame->pict_type == AV_PICTURE_TYPE_I) )
1968 self->first_pts = pts;
1969 if ( self->first_pts != AV_NOPTS_VALUE )
1970 pts -= self->first_pts;
1971 else if ( context->start_time != AV_NOPTS_VALUE )
1972 pts -= context->start_time;
1973 int_position = ( int64_t )( ( av_q2d( self->video_time_base ) * pts + delay ) * source_fps + 0.5 );
1974 }
1975
1976 if ( int_position < req_position )
1977 got_picture = 0;
1978 else if ( int_position >= req_position )
1979 codec_context->skip_loop_filter = AVDISCARD_NONE;
1980 }
1981 else if ( !self->pkt.data ) // draining decoder with null packets
1982 {
1983 self->video_send_result = -1;
1984 }
1985 mlt_log_debug( MLT_PRODUCER_SERVICE( producer ), " got_pic %d key %d send_result %d pkt_pos %"PRId64"\n",
1986 got_picture, self->pkt.flags & AV_PKT_FLAG_KEY, self->video_send_result, int_position );
1987 }
1988
1989 // Now handle the picture if we have one
1990 if ( got_picture )
1991 {
1992 #ifdef AVFILTER
1993 if (self->autorotate && self->vfilter_graph) {
1994 int ret = av_buffersrc_add_frame(self->vfilter_in, self->video_frame);
1995 if (ret < 0) {
1996 got_picture = 0;
1997 break;
1998 }
1999 while (ret >= 0) {
2000 ret = av_buffersink_get_frame_flags(self->vfilter_out, self->video_frame, 0);
2001 if (ret < 0) {
2002 ret = 0;
2003 break;
2004 }
2005 }
2006 }
2007 #endif
2008 set_image_size( self, width, height );
2009 if ( ( image_size = allocate_buffer( frame, codec_params, buffer, *format, *width, *height ) ) )
2010 {
2011 int yuv_colorspace;
2012 #if USE_HWACCEL
2013 // not sure why this is really needed, but doesn't seem to work otherwise
2014 yuv_colorspace = convert_image( self, self->video_frame, *buffer, self->video_frame->format,
2015 format, *width, *height, &alpha );
2016 #else
2017 yuv_colorspace = convert_image( self, self->video_frame, *buffer, codec_params->format,
2018 format, *width, *height, &alpha );
2019 #endif
2020 mlt_properties_set_int( frame_properties, "colorspace", yuv_colorspace );
2021 self->top_field_first |= self->video_frame->top_field_first;
2022 self->top_field_first |= codec_params->field_order == AV_FIELD_TT;
2023 self->top_field_first |= codec_params->field_order == AV_FIELD_TB;
2024 self->current_position = int_position;
2025 }
2026 else
2027 {
2028 got_picture = 0;
2029 }
2030 }
2031
2032 // Free packet data if not video and not live audio packet
2033 if ( self->pkt.stream_index != self->video_index &&
2034 !( !self->video_seekable && self->pkt.stream_index == self->audio_index ) )
2035 av_packet_unref( &self->pkt );
2036 }
2037 }
2038
2039 // set alpha
2040 if ( alpha )
2041 mlt_frame_set_alpha( frame, alpha, (*width) * (*height), mlt_pool_release );
2042
2043 if ( image_size > 0 )
2044 {
2045 mlt_properties_set_int( frame_properties, "format", *format );
2046 // Cache the image for rapid repeated access.
2047 if ( self->image_cache ) {
2048 if (is_album_art) {
2049 mlt_position original_pos = mlt_frame_original_position( frame );
2050 mlt_properties_set_position(frame_properties, "original_position", 0);
2051 mlt_cache_put_frame( self->image_cache, frame );
2052 mlt_properties_set_position(frame_properties, "original_position", original_pos);
2053 } else {
2054 mlt_cache_put_frame( self->image_cache, frame );
2055 }
2056 }
2057 // Clone frame for error concealment.
2058 if ( self->current_position >= self->last_good_position ) {
2059 self->last_good_position = self->current_position;
2060 if ( self->last_good_frame )
2061 mlt_frame_close( self->last_good_frame );
2062 self->last_good_frame = mlt_frame_clone( frame, 1 );
2063 }
2064 }
2065 else if ( self->last_good_frame )
2066 {
2067 // Use last known good frame if there was a decoding failure.
2068 mlt_frame original = mlt_frame_clone( self->last_good_frame, 1 );
2069 mlt_properties orig_props = MLT_FRAME_PROPERTIES( original );
2070 int size = 0;
2071
2072 *buffer = mlt_properties_get_data( orig_props, "alpha", &size );
2073 if (*buffer)
2074 mlt_frame_set_alpha( frame, *buffer, size, NULL );
2075 *buffer = mlt_properties_get_data( orig_props, "image", &size );
2076 mlt_frame_set_image( frame, *buffer, size, NULL );
2077 mlt_properties_set_data( frame_properties, "avformat.conceal_error", original, 0, (mlt_destructor) mlt_frame_close, NULL );
2078 *format = mlt_properties_get_int( orig_props, "format" );
2079 set_image_size( self, width, height );
2080 got_picture = 1;
2081 }
2082
2083 // Regardless of speed, we expect to get the next frame (cos we ain't too bright)
2084 self->video_expected = position + 1;
2085
2086 exit_get_image:
2087 pthread_mutex_unlock( &self->video_mutex );
2088
2089 // Set the progressive flag
2090 if ( mlt_properties_get( properties, "force_progressive" ) ) {
2091 mlt_properties_set_int( frame_properties, "progressive", !!mlt_properties_get_int( properties, "force_progressive" ) );
2092 } else if ( self->video_frame && codec_params ) {
2093 mlt_properties_set_int( frame_properties, "progressive",
2094 !self->video_frame->interlaced_frame &&
2095 (codec_params->field_order == AV_FIELD_PROGRESSIVE ||
2096 codec_params->field_order == AV_FIELD_UNKNOWN) );
2097 }
2098
2099 // Set the field order property for this frame
2100 if ( mlt_properties_get( properties, "force_tff" ) )
2101 mlt_properties_set_int( frame_properties, "top_field_first", !!mlt_properties_get_int( properties, "force_tff" ) );
2102 else
2103 mlt_properties_set_int( frame_properties, "top_field_first", self->top_field_first );
2104
2105 // Set immutable properties of the selected track's (or overridden) source attributes.
2106 mlt_properties_set_int( properties, "meta.media.top_field_first", self->top_field_first );
2107 mlt_properties_set_int( properties, "meta.media.progressive", mlt_properties_get_int( frame_properties, "progressive" ) );
2108 mlt_service_unlock( MLT_PRODUCER_SERVICE( producer ) );
2109
2110 mlt_log_timings_end( NULL, __FUNCTION__ );
2111
2112 return !got_picture;
2113 }
2114
2115 /** Process properties as AVOptions and apply to AV context obj
2116 */
2117
apply_properties(void * obj,mlt_properties properties,int flags)2118 static void apply_properties( void *obj, mlt_properties properties, int flags )
2119 {
2120 int i;
2121 int count = mlt_properties_count( properties );
2122 for ( i = 0; i < count; i++ )
2123 {
2124 const char *opt_name = mlt_properties_get_name( properties, i );
2125 int search_flags = AV_OPT_SEARCH_CHILDREN;
2126 const AVOption *opt = av_opt_find( obj, opt_name, NULL, flags, search_flags );
2127 if ( opt_name && mlt_properties_get( properties, opt_name ) && strcmp(opt_name, "seekable") )
2128 {
2129 if ( opt )
2130 av_opt_set( obj, opt_name, mlt_properties_get( properties, opt_name), search_flags );
2131 }
2132 }
2133 }
2134
2135 /** Initialize the video codec context.
2136 */
2137
video_codec_init(producer_avformat self,int index,mlt_properties properties)2138 static int video_codec_init( producer_avformat self, int index, mlt_properties properties )
2139 {
2140 // Initialise the codec if necessary
2141 if ( !self->video_codec )
2142 {
2143 // Get the video stream
2144 AVStream *stream = self->video_format->streams[ index ];
2145
2146 // Get codec context
2147 AVCodecContext *codec_context = stream->codec;
2148 AVCodecParameters *codec_params = stream->codecpar;
2149
2150 // Find the codec
2151 AVCodec *codec = avcodec_find_decoder( codec_params->codec_id );
2152 if ( mlt_properties_get( properties, "vcodec" ) ) {
2153 if ( !( codec = avcodec_find_decoder_by_name( mlt_properties_get( properties, "vcodec" ) ) ) )
2154 codec = avcodec_find_decoder( codec_params->codec_id );
2155 } else if ( codec_params->codec_id == AV_CODEC_ID_VP9 ) {
2156 if ( !( codec = avcodec_find_decoder_by_name( "libvpx-vp9" ) ) )
2157 codec = avcodec_find_decoder( codec_params->codec_id );
2158 } else if ( codec_params->codec_id == AV_CODEC_ID_VP8 ) {
2159 if ( !( codec = avcodec_find_decoder_by_name( "libvpx" ) ) )
2160 codec = avcodec_find_decoder( codec_params->codec_id );
2161 }
2162
2163 // Initialise multi-threading
2164 int thread_count = mlt_properties_get_int( properties, "threads" );
2165 if ( thread_count == 0 && getenv( "MLT_AVFORMAT_THREADS" ) )
2166 thread_count = atoi( getenv( "MLT_AVFORMAT_THREADS" ) );
2167 if ( thread_count >= 0 )
2168 codec_context->thread_count = thread_count;
2169
2170 #if USE_HWACCEL
2171 if ( self->hwaccel.device_type == AV_HWDEVICE_TYPE_NONE || self->hwaccel.pix_fmt == AV_PIX_FMT_NONE )
2172 {
2173 mlt_log_verbose( MLT_PRODUCER_SERVICE( self->parent ), "missing hwaccel parameters. skipping hardware initialization\n" );
2174 goto skip_hwaccel;
2175 }
2176
2177 int found_hw_pix_fmt = 0, i;
2178 for ( i = 0;; i++ )
2179 {
2180 const AVCodecHWConfig *config = avcodec_get_hw_config( codec, i );
2181 if ( !config )
2182 break;
2183
2184 if ( config->methods & AV_CODEC_HW_CONFIG_METHOD_HW_DEVICE_CTX &&
2185 config->device_type == self->hwaccel.device_type && config->pix_fmt == self->hwaccel.pix_fmt )
2186 {
2187 found_hw_pix_fmt = 1;
2188 break;
2189 }
2190 }
2191
2192 if ( found_hw_pix_fmt )
2193 {
2194 av_buffer_unref( &self->hwaccel.device_ctx );
2195 int ret = av_hwdevice_ctx_create( &self->hwaccel.device_ctx, self->hwaccel.device_type, self->hwaccel.device, NULL, 0 );
2196 if ( ret >= 0 )
2197 {
2198 codec_context->hw_device_ctx = av_buffer_ref( self->hwaccel.device_ctx );
2199 mlt_log_info( MLT_PRODUCER_SERVICE( self->parent ), "av_hwdevice_ctx_create() success %d\n", codec_context->pix_fmt );
2200 }
2201 else
2202 {
2203 mlt_log_warning( MLT_PRODUCER_SERVICE( self->parent ), "av_hwdevice_ctx_create() failed %d\n", ret );
2204 }
2205 }
2206 else
2207 {
2208 mlt_log_warning( MLT_PRODUCER_SERVICE( self->parent ), "failed to find hw_pix_fmt\n" );
2209 }
2210
2211 skip_hwaccel:
2212 #endif
2213 // If we don't have a codec and we can't initialise it, we can't do much more...
2214 pthread_mutex_lock( &self->open_mutex );
2215 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2216 {
2217 // Switch to the native vp8/vp9 decoder if not yuva420p
2218 if ( codec_params->format != AV_PIX_FMT_YUVA420P
2219 && !mlt_properties_get( properties, "vcodec" )
2220 && ( !strcmp(codec->name, "libvpx") || !strcmp(codec->name, "libvpx-vp9") ) )
2221 {
2222 codec = avcodec_find_decoder( codec_params->codec_id );
2223 if ( codec && avcodec_open2( codec_context, codec, NULL ) < 0 ) {
2224 self->video_index = -1;
2225 pthread_mutex_unlock( &self->open_mutex );
2226 return 0;
2227 }
2228 }
2229 // Now store the codec with its destructor
2230 self->video_codec = codec_context;
2231 }
2232 else
2233 {
2234 // Remember that we can't use this later
2235 self->video_index = -1;
2236 pthread_mutex_unlock( &self->open_mutex );
2237 return 0;
2238 }
2239 pthread_mutex_unlock( &self->open_mutex );
2240
2241 // Process properties as AVOptions
2242 apply_properties( codec_context, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2243 if ( codec && codec->priv_class && codec_context->priv_data )
2244 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_VIDEO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2245
2246 // Reset some image properties
2247 mlt_properties_set_int( properties, "width", codec_params->width );
2248 mlt_properties_set_int( properties, "height", codec_params->height );
2249 get_aspect_ratio( properties, stream, codec_params );
2250
2251 // Start with the muxer frame rate.
2252 AVRational frame_rate = stream->avg_frame_rate;
2253 double fps = av_q2d( frame_rate );
2254
2255 #if defined(FFUDIV)
2256 // Verify and sanitize the muxer frame rate.
2257 if ( isnan( fps ) || isinf( fps ) || fps == 0 )
2258 {
2259 frame_rate = stream->r_frame_rate;
2260 fps = av_q2d( frame_rate );
2261 }
2262 // With my samples when r_frame_rate != 1000 but avg_frame_rate is valid,
2263 // avg_frame_rate gives some approximate value that does not well match the media.
2264 // Also, on my sample where r_frame_rate = 1000, using avg_frame_rate directly
2265 // results in some very choppy output, but some value slightly different works
2266 // great.
2267 if ( av_q2d( stream->r_frame_rate ) >= 1000 && av_q2d( stream->avg_frame_rate ) > 0 )
2268 {
2269 frame_rate = av_d2q( av_q2d( stream->avg_frame_rate ), 1024 );
2270 fps = av_q2d( frame_rate );
2271 }
2272 #endif
2273 // XXX frame rates less than 1 fps are not considered sane
2274 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
2275 {
2276 // Get the frame rate from the codec.
2277 frame_rate.num = self->video_codec->time_base.den;
2278 frame_rate.den = self->video_codec->time_base.num * self->video_codec->ticks_per_frame;
2279 fps = av_q2d( frame_rate );
2280 }
2281 if ( isnan( fps ) || isinf( fps ) || fps < 1.0 )
2282 {
2283 // Use the profile frame rate if all else fails.
2284 mlt_profile profile = mlt_service_profile( MLT_PRODUCER_SERVICE( self->parent ) );
2285 frame_rate.num = profile->frame_rate_num;
2286 frame_rate.den = profile->frame_rate_den;
2287 }
2288
2289 self->video_time_base = stream->time_base;
2290 if ( mlt_properties_get( properties, "force_fps" ) )
2291 {
2292 AVRational force_fps = av_d2q( mlt_properties_get_double( properties, "force_fps" ), 1024 );
2293 self->video_time_base = av_mul_q( stream->time_base, av_div_q( frame_rate, force_fps ) );
2294 frame_rate = force_fps;
2295 }
2296 mlt_properties_set_int( properties, "meta.media.frame_rate_num", frame_rate.num );
2297 mlt_properties_set_int( properties, "meta.media.frame_rate_den", frame_rate.den );
2298
2299 // MP3 album art is a single JPEG at 90000 fps, which is not seekable.
2300 if ( codec->id == AV_CODEC_ID_MJPEG && frame_rate.num == 90000 && frame_rate.den == 1 )
2301 self->video_seekable = 0;
2302
2303 // Set the YUV colorspace from override or detect
2304 self->yuv_colorspace = mlt_properties_get_int( properties, "force_colorspace" );
2305 if ( ! self->yuv_colorspace )
2306 {
2307 switch ( self->video_codec->colorspace )
2308 {
2309 case AVCOL_SPC_SMPTE240M:
2310 self->yuv_colorspace = 240;
2311 break;
2312 case AVCOL_SPC_BT470BG:
2313 case AVCOL_SPC_SMPTE170M:
2314 self->yuv_colorspace = 601;
2315 break;
2316 case AVCOL_SPC_BT709:
2317 self->yuv_colorspace = 709;
2318 break;
2319 default:
2320 // This is a heuristic Charles Poynton suggests in "Digital Video and HDTV"
2321 self->yuv_colorspace = self->video_codec->width * self->video_codec->height > 750000 ? 709 : 601;
2322 break;
2323 }
2324 }
2325 // Let apps get chosen colorspace
2326 mlt_properties_set_int( properties, "meta.media.colorspace", self->yuv_colorspace );
2327
2328 // Get the color transfer characteristic (gamma).
2329 self->color_trc = mlt_properties_get_int( properties, "force_color_trc" );
2330 if ( !self->color_trc )
2331 self->color_trc = self->video_codec->color_trc;
2332 mlt_properties_set_int( properties, "meta.media.color_trc", self->color_trc );
2333
2334 // Get the RGB color primaries.
2335 switch ( self->video_codec->color_primaries )
2336 {
2337 case AVCOL_PRI_BT470BG:
2338 self->color_primaries = 601625;
2339 break;
2340 case AVCOL_PRI_SMPTE170M:
2341 case AVCOL_PRI_SMPTE240M:
2342 self->color_primaries = 601525;
2343 break;
2344 case AVCOL_PRI_BT709:
2345 case AVCOL_PRI_UNSPECIFIED:
2346 default:
2347 self->color_primaries = 709;
2348 break;
2349 }
2350
2351 self->full_luma = 0;
2352 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent), "color_range %d\n", codec_context->color_range );
2353 if ( codec_context->color_range == AVCOL_RANGE_JPEG )
2354 self->full_luma = 1;
2355 if ( mlt_properties_get( properties, "set.force_full_luma" ) )
2356 self->full_luma = mlt_properties_get_int( properties, "set.force_full_luma" );
2357 }
2358 return self->video_index > -1;
2359 }
2360
2361 /** Set up video handling.
2362 */
2363
producer_set_up_video(producer_avformat self,mlt_frame frame)2364 static void producer_set_up_video( producer_avformat self, mlt_frame frame )
2365 {
2366 // Get the producer
2367 mlt_producer producer = self->parent;
2368
2369 // Get the properties
2370 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2371
2372 // Fetch the video format context
2373 AVFormatContext *context = self->video_format;
2374
2375 // Get the video_index
2376 int index = mlt_properties_get_int( properties, "video_index" );
2377
2378 int unlock_needed = 0;
2379
2380 // Reopen the file if necessary
2381 if ( !context && index > -1 )
2382 {
2383 unlock_needed = 1;
2384 pthread_mutex_lock( &self->video_mutex );
2385 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2386 mlt_properties_get( properties, "resource" ), 0, 0 );
2387 context = self->video_format;
2388 }
2389
2390 // Exception handling for video_index
2391 if ( context && index >= (int) context->nb_streams )
2392 {
2393 // Get the last video stream
2394 for ( index = context->nb_streams - 1;
2395 index >= 0 && context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_VIDEO;
2396 index-- );
2397 mlt_properties_set_int( properties, "video_index", index );
2398 }
2399 if ( context && index > -1 && context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_VIDEO )
2400 {
2401 // Invalidate the video stream
2402 index = -1;
2403 mlt_properties_set_int( properties, "video_index", index );
2404 }
2405
2406 // Update the video properties if the index changed
2407 if ( context && index > -1 && index != self->video_index )
2408 {
2409 // Reset the video properties if the index changed
2410 self->video_index = index;
2411 pthread_mutex_lock( &self->open_mutex );
2412 if ( self->video_codec )
2413 avcodec_close( self->video_codec );
2414 self->video_codec = NULL;
2415 pthread_mutex_unlock( &self->open_mutex );
2416 }
2417
2418 // Get the frame properties
2419 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2420
2421 // Get the codec
2422 if ( context && index > -1 && video_codec_init( self, index, properties ) )
2423 {
2424 // Set the frame properties
2425 double force_aspect_ratio = mlt_properties_get_double( properties, "force_aspect_ratio" );
2426 double aspect_ratio = ( force_aspect_ratio > 0.0 ) ?
2427 force_aspect_ratio : mlt_properties_get_double( properties, "aspect_ratio" );
2428
2429 // Set the width and height
2430 double dar = mlt_profile_dar( mlt_service_profile( MLT_PRODUCER_SERVICE( producer ) ) );
2431 double theta = self->autorotate? get_rotation( self->video_format->streams[index] ) : 0.0;
2432 if ( fabs(theta - 90.0) < 1.0 || fabs(theta - 270.0) < 1.0 )
2433 {
2434 // Workaround 1088 encodings missing cropping info.
2435 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 ) {
2436 mlt_properties_set_int( frame_properties, "width", 1080 );
2437 mlt_properties_set_int( properties, "meta.media.width", 1080 );
2438 } else {
2439 mlt_properties_set_int( frame_properties, "width", self->video_codec->height );
2440 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->height );
2441 }
2442 mlt_properties_set_int( frame_properties, "height", self->video_codec->width );
2443 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->width );
2444 aspect_ratio = ( force_aspect_ratio > 0.0 ) ? force_aspect_ratio : 1.0 / aspect_ratio;
2445 mlt_properties_set_double( frame_properties, "aspect_ratio", 1.0/aspect_ratio );
2446 } else {
2447 mlt_properties_set_int( frame_properties, "width", self->video_codec->width );
2448 mlt_properties_set_int( properties, "meta.media.width", self->video_codec->width );
2449 // Workaround 1088 encodings missing cropping info.
2450 if ( self->video_codec->height == 1088 && dar == 16.0/9.0 ) {
2451 mlt_properties_set_int( frame_properties, "height", 1080 );
2452 mlt_properties_set_int( properties, "meta.media.height", 1080 );
2453 } else {
2454 mlt_properties_set_int( frame_properties, "height", self->video_codec->height );
2455 mlt_properties_set_int( properties, "meta.media.height", self->video_codec->height );
2456 }
2457 mlt_properties_set_double( frame_properties, "aspect_ratio", aspect_ratio );
2458 }
2459 mlt_properties_set_int( frame_properties, "colorspace", self->yuv_colorspace );
2460 mlt_properties_set_int( frame_properties, "color_trc", self->color_trc );
2461 mlt_properties_set_int( frame_properties, "color_primaries", self->color_primaries );
2462 mlt_properties_set_int( frame_properties, "full_luma", self->full_luma );
2463 mlt_properties_set( properties, "meta.media.color_range", self->full_luma? "full" : "mpeg" );
2464
2465 // Add our image operation
2466 mlt_frame_push_service( frame, self );
2467 mlt_frame_push_get_image( frame, producer_get_image );
2468 }
2469 else
2470 {
2471 // If something failed, use test card image
2472 mlt_properties_set_int( frame_properties, "test_image", 1 );
2473 }
2474 if ( unlock_needed )
2475 pthread_mutex_unlock( &self->video_mutex );
2476 }
2477
seek_audio(producer_avformat self,mlt_position position,double timecode)2478 static int seek_audio( producer_avformat self, mlt_position position, double timecode )
2479 {
2480 int paused = 0;
2481
2482 pthread_mutex_lock( &self->packets_mutex );
2483
2484 // Seek if necessary
2485 if ( self->seekable && ( position != self->audio_expected || self->last_position < 0 ) )
2486 {
2487 if ( self->last_position == POSITION_INITIAL )
2488 {
2489 int video_index = self->video_index;
2490 if ( video_index == -1 )
2491 video_index = first_video_index( self );
2492 if ( self->first_pts == AV_NOPTS_VALUE && video_index >= 0 )
2493 find_first_pts( self, video_index );
2494 }
2495
2496 if ( position + 1 == self->audio_expected &&
2497 mlt_properties_get_int( MLT_PRODUCER_PROPERTIES( self->parent ), "mute_on_pause" ) )
2498 {
2499 // We're paused - silence required
2500 paused = 1;
2501 }
2502 else if ( position < self->audio_expected || position - self->audio_expected >= 12 )
2503 {
2504 AVFormatContext *context = self->audio_format;
2505 int64_t timestamp = llrint( timecode * AV_TIME_BASE );
2506 if ( context->start_time != AV_NOPTS_VALUE )
2507 timestamp += context->start_time;
2508 if ( timestamp < 0 )
2509 timestamp = 0;
2510
2511 // Set to the real timecode
2512 if ( av_seek_frame( context, -1, timestamp, AVSEEK_FLAG_BACKWARD ) != 0 )
2513 paused = 1;
2514
2515 // Clear the usage in the audio buffer
2516 int i = MAX_AUDIO_STREAMS + 1;
2517 while ( --i )
2518 self->audio_used[i - 1] = 0;
2519 }
2520 }
2521 pthread_mutex_unlock( &self->packets_mutex );
2522 return paused;
2523 }
2524
sample_bytes(AVCodecContext * context)2525 static int sample_bytes( AVCodecContext *context )
2526 {
2527 return av_get_bytes_per_sample( context->sample_fmt );
2528 }
2529
planar_to_interleaved(uint8_t * dest,AVFrame * src,int samples,int channels,int bytes_per_sample)2530 static void planar_to_interleaved( uint8_t *dest, AVFrame *src, int samples, int channels, int bytes_per_sample )
2531 {
2532 int s, c;
2533 for ( s = 0; s < samples; s++ )
2534 {
2535 for ( c = 0; c < channels; c++ )
2536 {
2537 if ( c < AV_NUM_DATA_POINTERS )
2538 memcpy( dest, &src->data[c][s * bytes_per_sample], bytes_per_sample );
2539 dest += bytes_per_sample;
2540 }
2541 }
2542 }
2543
decode_audio(producer_avformat self,int * ignore,const AVPacket * pkt,int samples,double timecode,double fps)2544 static int decode_audio( producer_avformat self, int *ignore, const AVPacket *pkt, int samples, double timecode, double fps )
2545 {
2546 // Fetch the audio_format
2547 AVFormatContext *context = self->audio_format;
2548
2549 // Get the current stream index
2550 int index = pkt->stream_index;
2551
2552 // Get codec context
2553 AVCodecContext *codec_context = self->audio_codec[ index ];
2554
2555 // Obtain the audio buffers
2556 uint8_t *audio_buffer = self->audio_buffer[ index ];
2557
2558 int channels = codec_context->channels;
2559 int audio_used = self->audio_used[ index ];
2560 int ret = 0;
2561 int discarded = 1;
2562 int sizeof_sample = sample_bytes( codec_context );
2563
2564 // Decode the audio
2565 if ( !self->audio_frame )
2566 self->audio_frame = av_frame_alloc();
2567 else
2568 av_frame_unref( self->audio_frame );
2569 int error = avcodec_send_packet(codec_context, pkt);
2570 mlt_log_debug(MLT_PRODUCER_SERVICE(self->parent), "decoded audio packet with size %d => %d\n", pkt->size, error);
2571 if (!ignore_send_packet_result(error)) {
2572 mlt_log_warning(MLT_PRODUCER_SERVICE(self->parent), "audio avcodec_send_packet failed with %d\n", error);
2573 } else while (!error) {
2574 error = avcodec_receive_frame(codec_context, self->audio_frame);
2575 if (error) {
2576 if (error != AVERROR(EAGAIN)) {
2577 mlt_log_warning(MLT_PRODUCER_SERVICE(self->parent), "audio decoding error %d\n", error);
2578 }
2579 } else {
2580 // Figure out how many samples will be needed after resampling
2581 int convert_samples = self->audio_frame->nb_samples;
2582 channels = codec_context->channels;
2583 ret += convert_samples * channels * sizeof_sample;
2584
2585 // Resize audio buffer to prevent overflow
2586 if ( ( audio_used + convert_samples ) * channels * sizeof_sample > self->audio_buffer_size[ index ] )
2587 {
2588 self->audio_buffer_size[ index ] = ( audio_used + convert_samples * 2 ) * channels * sizeof_sample;
2589 audio_buffer = self->audio_buffer[ index ] = mlt_pool_realloc( audio_buffer, self->audio_buffer_size[ index ] );
2590 }
2591 uint8_t *dest = &audio_buffer[ audio_used * channels * sizeof_sample ];
2592 switch ( codec_context->sample_fmt )
2593 {
2594 case AV_SAMPLE_FMT_U8P:
2595 case AV_SAMPLE_FMT_S16P:
2596 case AV_SAMPLE_FMT_S32P:
2597 case AV_SAMPLE_FMT_FLTP:
2598 planar_to_interleaved( dest, self->audio_frame, convert_samples, channels, sizeof_sample );
2599 break;
2600 default: {
2601 int data_size = av_samples_get_buffer_size( NULL, channels,
2602 self->audio_frame->nb_samples, codec_context->sample_fmt, 1 );
2603 // Straight copy to audio buffer
2604 memcpy( dest, self->audio_frame->data[0], data_size );
2605 }
2606 }
2607 audio_used += convert_samples;
2608 discarded = 0;
2609 }
2610 }
2611
2612 // Handle ignore
2613 if ( *ignore > 0 && audio_used )
2614 {
2615 int n = FFMIN( audio_used, *ignore );
2616 *ignore -= n;
2617 audio_used -= n;
2618 memmove( audio_buffer, &audio_buffer[ n * channels * sizeof_sample ],
2619 audio_used * channels * sizeof_sample );
2620 }
2621
2622 // If we're behind, ignore this packet
2623 // Skip this on non-seekable, audio-only inputs.
2624 if ( !discarded && pkt->pts >= 0 && ( self->seekable || self->video_format ) && *ignore == 0 && audio_used > samples / 2 )
2625 {
2626 int64_t pts = pkt->pts;
2627 if ( self->first_pts != AV_NOPTS_VALUE )
2628 pts -= self->first_pts;
2629 else if ( context->start_time != AV_NOPTS_VALUE && self->video_index != -1 )
2630 pts -= context->start_time;
2631 double timebase = av_q2d( context->streams[ index ]->time_base );
2632 int64_t int_position = llrint( timebase * pts * fps );
2633 int64_t req_position = llrint( timecode * fps );
2634 int64_t req_pts = llrint( timecode / timebase );
2635
2636 mlt_log_debug( MLT_PRODUCER_SERVICE(self->parent),
2637 "A pkt.pts %"PRId64" pkt->dts %"PRId64" req_pos %"PRId64" cur_pos %"PRId64" pkt_pos %"PRId64"\n",
2638 pkt->pts, pkt->dts, req_position, self->current_position, int_position );
2639
2640 if ( self->seekable || int_position > 0 )
2641 {
2642 if ( req_position > int_position ) {
2643 // We are behind, so skip some
2644 *ignore = lrint( timebase * (req_pts - pts) * codec_context->sample_rate );
2645 } else if ( self->audio_index != INT_MAX && int_position > req_position + 2 && !self->is_audio_synchronizing ) {
2646 // We are ahead, so seek backwards some more.
2647 // Supply -1 as the position to defeat the checks needed by for the other
2648 // call to seek_audio() at the beginning of producer_get_audio(). Otherwise,
2649 // more often than not, req_position will equal audio_expected.
2650 seek_audio( self, -1, timecode - 1.0 );
2651 self->is_audio_synchronizing = 1;
2652 }
2653 }
2654
2655 // Cancel the find_first_pts() in seek_audio()
2656 if ( self->video_index == -1 && self->last_position == POSITION_INITIAL )
2657 self->last_position = int_position;
2658 }
2659
2660 self->audio_used[ index ] = audio_used;
2661
2662 return ret;
2663 }
2664
2665 /** Get the audio from a frame.
2666 */
producer_get_audio(mlt_frame frame,void ** buffer,mlt_audio_format * format,int * frequency,int * channels,int * samples)2667 static int producer_get_audio( mlt_frame frame, void **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
2668 {
2669 // Get the producer
2670 producer_avformat self = mlt_frame_pop_audio( frame );
2671
2672 pthread_mutex_lock( &self->audio_mutex );
2673
2674 // Obtain the frame number of this frame
2675 mlt_position position = mlt_frame_original_position( frame );
2676
2677 // Calculate the real time code
2678 double real_timecode = producer_time_of_frame( self->parent, position );
2679
2680 // Get the producer fps
2681 double fps = mlt_producer_get_fps( self->parent );
2682 if ( mlt_properties_get( MLT_FRAME_PROPERTIES(frame), "producer_consumer_fps" ) )
2683 fps = mlt_properties_get_double( MLT_FRAME_PROPERTIES(frame), "producer_consumer_fps" );
2684
2685 // Number of frames to ignore (for ffwd)
2686 int ignore[ MAX_AUDIO_STREAMS ] = { 0 };
2687
2688 // Flag for paused (silence)
2689 double timecode = self->audio_expected > 0 ? real_timecode : FFMAX(real_timecode - 0.25, 0.0);
2690 int paused = seek_audio( self, position, timecode );
2691
2692 // Initialize ignore for all streams from the seek return value
2693 int i = MAX_AUDIO_STREAMS;
2694 while ( i-- )
2695 ignore[i] = ignore[0];
2696
2697 // Fetch the audio_format
2698 AVFormatContext *context = self->audio_format;
2699 if ( !context )
2700 goto exit_get_audio;
2701
2702 int sizeof_sample = sizeof( int16_t );
2703
2704 // Determine the tracks to use
2705 int index = self->audio_index;
2706 int index_max = self->audio_index + 1;
2707 if ( self->audio_index == INT_MAX )
2708 {
2709 index = 0;
2710 index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
2711 *channels = self->total_channels;
2712 *samples = mlt_audio_calculate_frame_samples( fps, self->max_frequency, position );
2713 *frequency = self->max_frequency;
2714 }
2715
2716 // Initialize the buffers
2717 for ( ; index < index_max && index < MAX_AUDIO_STREAMS; index++ )
2718 {
2719 // Get codec context
2720 AVCodecContext *codec_context = self->audio_codec[ index ];
2721
2722 if ( codec_context && !self->audio_buffer[ index ] )
2723 {
2724 if ( self->audio_index != INT_MAX && !mlt_properties_get( MLT_PRODUCER_PROPERTIES(self->parent), "request_channel_layout" ) )
2725 codec_context->request_channel_layout = av_get_default_channel_layout( *channels );
2726 sizeof_sample = sample_bytes( codec_context );
2727
2728 // Check for audio buffer and create if necessary
2729 self->audio_buffer_size[ index ] = MAX_AUDIO_FRAME_SIZE * sizeof_sample;
2730 self->audio_buffer[ index ] = mlt_pool_alloc( self->audio_buffer_size[ index ] );
2731
2732 // Check for decoder buffer and create if necessary
2733 self->decode_buffer[ index ] = av_malloc( self->audio_buffer_size[ index ] );
2734 }
2735 }
2736
2737 // Get the audio if required
2738 if ( !paused && *frequency > 0 )
2739 {
2740 int ret = 0;
2741 int got_audio = 0;
2742 AVPacket pkt;
2743 mlt_channel_layout layout = mlt_channel_auto;
2744
2745 av_init_packet( &pkt );
2746
2747 // Caller requested number samples based on requested sample rate.
2748 if ( self->audio_index != INT_MAX )
2749 *samples = mlt_audio_calculate_frame_samples( fps, self->audio_codec[ self->audio_index ]->sample_rate, position );
2750
2751 while ( ret >= 0 && !got_audio )
2752 {
2753 // Check if the buffer already contains the samples required
2754 if ( self->audio_index != INT_MAX &&
2755 self->audio_used[ self->audio_index ] >= *samples &&
2756 ignore[ self->audio_index ] == 0 )
2757 {
2758 got_audio = 1;
2759 break;
2760 }
2761 else if ( self->audio_index == INT_MAX )
2762 {
2763 // Check if there is enough audio for all streams
2764 got_audio = 1;
2765 for ( index = 0; got_audio && index < index_max; index++ )
2766 if ( ( self->audio_codec[ index ] && self->audio_used[ index ] < *samples ) || ignore[ index ] )
2767 got_audio = 0;
2768 if ( got_audio )
2769 break;
2770 }
2771
2772 // Read a packet
2773 pthread_mutex_lock( &self->packets_mutex );
2774 if ( mlt_deque_count( self->apackets ) )
2775 {
2776 AVPacket *tmp = (AVPacket*) mlt_deque_pop_front( self->apackets );
2777 av_packet_ref( &pkt, tmp );
2778 av_packet_free( &tmp );
2779 }
2780 else
2781 {
2782 ret = av_read_frame( context, &pkt );
2783 if ( ret >= 0 && !self->seekable && pkt.stream_index == self->video_index )
2784 {
2785 mlt_deque_push_back( self->vpackets, av_packet_clone(&pkt) );
2786 }
2787 else if ( ret < 0 )
2788 {
2789 mlt_producer producer = self->parent;
2790 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2791 if ( ret != AVERROR_EOF )
2792 mlt_log_verbose( MLT_PRODUCER_SERVICE(producer), "av_read_frame returned error %d inside get_audio\n", ret );
2793 if ( !self->seekable && mlt_properties_get_int( properties, "reconnect" ) )
2794 {
2795 // Try to reconnect to live sources by closing context and codecs,
2796 // and letting next call to get_frame() reopen.
2797 prepare_reopen( self );
2798 pthread_mutex_unlock( &self->packets_mutex );
2799 goto exit_get_audio;
2800 }
2801 if ( !self->seekable && mlt_properties_get_int( properties, "exit_on_disconnect" ) )
2802 {
2803 mlt_log_fatal( MLT_PRODUCER_SERVICE(producer), "Exiting with error due to disconnected source.\n" );
2804 exit( EXIT_FAILURE );
2805 }
2806 }
2807 }
2808 pthread_mutex_unlock( &self->packets_mutex );
2809
2810 // We only deal with audio from the selected audio index
2811 index = pkt.stream_index;
2812 if ( index < MAX_AUDIO_STREAMS && ret >= 0 && pkt.data && pkt.size > 0 && ( index == self->audio_index ||
2813 ( self->audio_index == INT_MAX && context->streams[ index ]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO ) ) )
2814 {
2815 ret = decode_audio( self, &ignore[index], &pkt, *samples, real_timecode, fps );
2816 }
2817
2818 if ( self->seekable || index != self->video_index )
2819 av_packet_unref( &pkt );
2820 }
2821 self->is_audio_synchronizing = 0;
2822
2823 // Set some additional return values
2824 *format = mlt_audio_s16;
2825 if ( self->audio_index != INT_MAX )
2826 {
2827 index = self->audio_index;
2828 *channels = self->audio_codec[ index ]->channels;
2829 *frequency = self->audio_codec[ index ]->sample_rate;
2830 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2831 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2832 if( self->audio_codec[ index ]->channel_layout == 0 )
2833 layout = av_channel_layout_to_mlt( av_get_default_channel_layout( self->audio_codec[ index ]->channels ) );
2834 else
2835 layout = av_channel_layout_to_mlt( self->audio_codec[ index ]->channel_layout );
2836 }
2837 else if ( self->audio_index == INT_MAX )
2838 {
2839 layout = mlt_channel_independent;
2840 for ( index = 0; index < index_max; index++ )
2841 if ( self->audio_codec[ index ] )
2842 {
2843 // XXX: This only works if all audio tracks have the same sample format.
2844 *format = pick_audio_format( self->audio_codec[ index ]->sample_fmt );
2845 sizeof_sample = sample_bytes( self->audio_codec[ index ] );
2846 break;
2847 }
2848 }
2849 mlt_properties_set( MLT_FRAME_PROPERTIES(frame), "channel_layout", mlt_audio_channel_layout_name( layout ) );
2850
2851 // Allocate and set the frame's audio buffer
2852 int size = mlt_audio_format_size( *format, *samples, *channels );
2853 *buffer = mlt_pool_alloc( size );
2854 mlt_frame_set_audio( frame, *buffer, *format, size, mlt_pool_release );
2855
2856 // Interleave tracks if audio_index=all
2857 if ( self->audio_index == INT_MAX )
2858 {
2859 uint8_t *dest = *buffer;
2860 int i;
2861 for ( i = 0; i < *samples; i++ )
2862 {
2863 for ( index = 0; index < index_max; index++ )
2864 if ( self->audio_codec[ index ] )
2865 {
2866 int current_channels = self->audio_codec[ index ]->channels;
2867 uint8_t *src = self->audio_buffer[ index ] + i * current_channels * sizeof_sample;
2868 memcpy( dest, src, current_channels * sizeof_sample );
2869 dest += current_channels * sizeof_sample;
2870 }
2871 }
2872 for ( index = 0; index < index_max; index++ )
2873 if ( self->audio_codec[ index ] && self->audio_used[ index ] >= *samples )
2874 {
2875 int current_channels = self->audio_codec[ index ]->channels;
2876 uint8_t *src = self->audio_buffer[ index ] + *samples * current_channels * sizeof_sample;
2877 self->audio_used[index] -= *samples;
2878 memmove( self->audio_buffer[ index ], src, self->audio_used[ index ] * current_channels * sizeof_sample );
2879 }
2880 }
2881 // Copy a single track to the output buffer
2882 else
2883 {
2884 index = self->audio_index;
2885 uint8_t silence = *format == mlt_audio_u8 ? 0x80 : 0;
2886
2887 // Now handle the audio if we have enough
2888 if ( self->audio_used[ index ] > 0 )
2889 {
2890 uint8_t *src = self->audio_buffer[ index ];
2891 // copy samples from audio_buffer
2892 size = self->audio_used[ index ] < *samples ? self->audio_used[ index ] : *samples;
2893 memcpy( *buffer, src, size * *channels * sizeof_sample );
2894 // supply the remaining requested samples as silence
2895 if ( *samples > self->audio_used[ index ] )
2896 memset( *buffer + size * *channels * sizeof_sample, silence, ( *samples - self->audio_used[ index ] ) * *channels * sizeof_sample );
2897 // reposition the samples within audio_buffer
2898 self->audio_used[ index ] -= size;
2899 memmove( src, src + size * *channels * sizeof_sample, self->audio_used[ index ] * *channels * sizeof_sample );
2900 }
2901 else
2902 {
2903 // Otherwise fill with silence
2904 memset( *buffer, silence, *samples * *channels * sizeof_sample );
2905 }
2906 }
2907 }
2908 else
2909 {
2910 exit_get_audio:
2911 // Get silence and don't touch the context
2912 mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
2913 }
2914
2915 // Regardless of speed (other than paused), we expect to get the next frame
2916 if ( !paused )
2917 self->audio_expected = position + 1;
2918
2919 pthread_mutex_unlock( &self->audio_mutex );
2920
2921 return 0;
2922 }
2923
2924 /** Initialize the audio codec context.
2925 */
2926
audio_codec_init(producer_avformat self,int index,mlt_properties properties)2927 static int audio_codec_init( producer_avformat self, int index, mlt_properties properties )
2928 {
2929 // Initialise the codec if necessary
2930 if ( !self->audio_codec[ index ] )
2931 {
2932 // Get codec context
2933 AVCodecContext *codec_context = self->audio_format->streams[index]->codec;
2934
2935 // Find the codec
2936 AVCodec *codec = avcodec_find_decoder( codec_context->codec_id );
2937 if ( mlt_properties_get( properties, "acodec" ) )
2938 {
2939 if ( !( codec = avcodec_find_decoder_by_name( mlt_properties_get( properties, "acodec" ) ) ) )
2940 codec = avcodec_find_decoder( codec_context->codec_id );
2941 }
2942
2943 // If we don't have a codec and we can't initialise it, we can't do much more...
2944 pthread_mutex_lock( &self->open_mutex );
2945 if ( codec && avcodec_open2( codec_context, codec, NULL ) >= 0 )
2946 {
2947 // Now store the codec with its destructor
2948 if ( self->audio_codec[ index ] )
2949 avcodec_close( self->audio_codec[ index ] );
2950 self->audio_codec[ index ] = codec_context;
2951 self->audio_index = index;
2952 }
2953 else
2954 {
2955 // Remember that we can't use self later
2956 self->audio_index = -1;
2957 }
2958 pthread_mutex_unlock( &self->open_mutex );
2959
2960 // Process properties as AVOptions
2961 apply_properties( codec_context, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2962 if ( codec && codec->priv_class && codec_context->priv_data )
2963 apply_properties( codec_context->priv_data, properties, AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM );
2964 }
2965 return self->audio_codec[ index ] && self->audio_index > -1;
2966 }
2967
2968 /** Set up audio handling.
2969 */
2970
producer_set_up_audio(producer_avformat self,mlt_frame frame)2971 static void producer_set_up_audio( producer_avformat self, mlt_frame frame )
2972 {
2973 // Get the producer
2974 mlt_producer producer = self->parent;
2975
2976 // Get the properties
2977 mlt_properties properties = MLT_PRODUCER_PROPERTIES( producer );
2978
2979 // Fetch the audio format context
2980 AVFormatContext *context = self->audio_format;
2981
2982 mlt_properties frame_properties = MLT_FRAME_PROPERTIES( frame );
2983
2984 // Get the audio_index
2985 int index = mlt_properties_get_int( properties, "audio_index" );
2986
2987 // Handle all audio tracks
2988 if ( self->audio_index > -1 &&
2989 mlt_properties_get( properties, "audio_index" ) &&
2990 !strcmp( mlt_properties_get( properties, "audio_index" ), "all" ) )
2991 index = INT_MAX;
2992
2993 // Reopen the file if necessary
2994 if ( !context && self->audio_index > -1 && index > -1 )
2995 {
2996 producer_open( self, mlt_service_profile( MLT_PRODUCER_SERVICE(producer) ),
2997 mlt_properties_get( properties, "resource" ), 1, 0 );
2998 context = self->audio_format;
2999 }
3000
3001 // Exception handling for audio_index
3002 if ( context && index >= (int) context->nb_streams && index < INT_MAX )
3003 {
3004 for ( index = context->nb_streams - 1;
3005 index >= 0 && context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_AUDIO;
3006 index-- );
3007 mlt_properties_set_int( properties, "audio_index", index );
3008 }
3009 if ( context && index > -1 && index < INT_MAX &&
3010 context->streams[ index ]->codecpar->codec_type != AVMEDIA_TYPE_AUDIO )
3011 {
3012 index = self->audio_index;
3013 mlt_properties_set_int( properties, "audio_index", index );
3014 }
3015 if ( context && index > -1 && index < INT_MAX &&
3016 pick_audio_format( context->streams[ index ]->codecpar->format ) == mlt_audio_none )
3017 {
3018 index = -1;
3019 }
3020
3021 // Update the audio properties if the index changed
3022 if ( context && self->audio_index > -1 && index != self->audio_index )
3023 {
3024 pthread_mutex_lock( &self->open_mutex );
3025 unsigned i = 0;
3026 int index_max = FFMIN( MAX_AUDIO_STREAMS, context->nb_streams );
3027 for (i = 0; i < index_max; i++) {
3028 if (self->audio_codec[i]) {
3029 avcodec_close(self->audio_codec[i]);
3030 self->audio_codec[i] = NULL;
3031 }
3032 }
3033 pthread_mutex_unlock( &self->open_mutex );
3034 }
3035
3036 // Get the codec(s)
3037 if ( context && index == INT_MAX )
3038 {
3039 unsigned int index;
3040 mlt_properties_set_int( frame_properties, "audio_frequency", self->max_frequency );
3041 mlt_properties_set_int( frame_properties, "audio_channels", self->total_channels );
3042 for ( index = 0; index < context->nb_streams && index < MAX_AUDIO_STREAMS; index++ )
3043 {
3044 if ( context->streams[ index ]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO )
3045 audio_codec_init( self, index, properties );
3046 }
3047 self->audio_index = INT_MAX;
3048 }
3049 else if ( context && index > -1 && index < MAX_AUDIO_STREAMS &&
3050 audio_codec_init( self, index, properties ) )
3051 {
3052 mlt_properties_set_int( frame_properties, "audio_frequency", self->audio_codec[ index ]->sample_rate );
3053 mlt_properties_set_int( frame_properties, "audio_channels", self->audio_codec[ index ]->channels );
3054 }
3055 if ( context && index > -1 )
3056 {
3057 // Add our audio operation
3058 mlt_frame_push_audio( frame, self );
3059 mlt_frame_push_audio( frame, producer_get_audio );
3060 }
3061 }
3062
3063 /** Our get frame implementation.
3064 */
3065
producer_get_frame(mlt_producer producer,mlt_frame_ptr frame,int index)3066 static int producer_get_frame( mlt_producer producer, mlt_frame_ptr frame, int index )
3067 {
3068 // Access the private data
3069 (void) index; // unused
3070 mlt_service service = MLT_PRODUCER_SERVICE( producer );
3071 mlt_cache_item cache_item = mlt_service_cache_get( service, "producer_avformat" );
3072 producer_avformat self = mlt_cache_item_data( cache_item, NULL );
3073
3074 // If cache miss
3075 if ( !self )
3076 {
3077 self = calloc( 1, sizeof( struct producer_avformat_s ) );
3078 producer->child = self;
3079 self->parent = producer;
3080 mlt_service_cache_put( service, "producer_avformat", self, 0, (mlt_destructor) producer_avformat_close );
3081 cache_item = mlt_service_cache_get( service, "producer_avformat" );
3082 }
3083
3084 // Create an empty frame
3085 *frame = mlt_frame_init( service);
3086
3087 if ( *frame )
3088 {
3089 mlt_properties_set_data( MLT_FRAME_PROPERTIES(*frame), "avformat_cache", cache_item, 0, (mlt_destructor) mlt_cache_item_close, NULL );
3090 }
3091 else
3092 {
3093 mlt_cache_item_close( cache_item );
3094 return 1;
3095 }
3096
3097 // Update timecode on the frame we're creating
3098 mlt_frame_set_position( *frame, mlt_producer_position( producer ) );
3099
3100 // Set up the video
3101 producer_set_up_video( self, *frame );
3102
3103 // Set up the audio
3104 producer_set_up_audio( self, *frame );
3105
3106 // Set the position of this producer
3107 mlt_position position = mlt_producer_frame( producer );
3108 mlt_properties_set_position( MLT_FRAME_PROPERTIES( *frame ), "original_position", position );
3109
3110 // Calculate the next timecode
3111 mlt_producer_prepare_next( producer );
3112
3113 return 0;
3114 }
3115
producer_avformat_close(producer_avformat self)3116 static void producer_avformat_close( producer_avformat self )
3117 {
3118 mlt_log_debug( NULL, "producer_avformat_close\n" );
3119
3120 // Cleanup av contexts
3121 av_packet_unref( &self->pkt );
3122 av_frame_free( &self->video_frame );
3123 av_frame_free( &self->audio_frame );
3124
3125 #if USE_HWACCEL
3126 av_buffer_unref( &self->hwaccel.device_ctx );
3127 #endif
3128
3129 if ( self->is_mutex_init )
3130 pthread_mutex_lock( &self->open_mutex );
3131 int i;
3132 for ( i = 0; i < MAX_AUDIO_STREAMS; i++ )
3133 {
3134 mlt_pool_release( self->audio_buffer[i] );
3135 av_free( self->decode_buffer[i] );
3136 if ( self->audio_codec[i] )
3137 avcodec_close( self->audio_codec[i] );
3138 self->audio_codec[i] = NULL;
3139 }
3140 if ( self->video_codec )
3141 avcodec_close( self->video_codec );
3142 self->video_codec = NULL;
3143 // Close the file
3144 if ( self->dummy_context )
3145 avformat_close_input( &self->dummy_context );
3146 if ( self->seekable && self->audio_format )
3147 avformat_close_input( &self->audio_format );
3148 if ( self->video_format )
3149 avformat_close_input( &self->video_format );
3150 if ( self->is_mutex_init )
3151 pthread_mutex_unlock( &self->open_mutex );
3152 #ifdef AVFILTER
3153 avfilter_graph_free(&self->vfilter_graph);
3154 #endif
3155
3156 // Cleanup caches.
3157 mlt_cache_close( self->image_cache );
3158 if ( self->last_good_frame )
3159 mlt_frame_close( self->last_good_frame );
3160
3161 // Cleanup the mutexes
3162 if ( self->is_mutex_init )
3163 {
3164 pthread_mutex_destroy( &self->audio_mutex );
3165 pthread_mutex_destroy( &self->video_mutex );
3166 pthread_mutex_destroy( &self->packets_mutex );
3167 pthread_mutex_destroy( &self->open_mutex );
3168 }
3169
3170 // Cleanup the packet queues
3171 AVPacket *pkt;
3172 if ( self->apackets )
3173 {
3174 while ( ( pkt = mlt_deque_pop_back( self->apackets ) ) )
3175 {
3176 av_packet_free( &pkt );
3177 }
3178 mlt_deque_close( self->apackets );
3179 self->apackets = NULL;
3180 }
3181 if ( self->vpackets )
3182 {
3183 while ( ( pkt = mlt_deque_pop_back( self->vpackets ) ) )
3184 {
3185 av_packet_free( &pkt );
3186 }
3187 mlt_deque_close( self->vpackets );
3188 self->vpackets = NULL;
3189 }
3190
3191 free( self );
3192 }
3193
producer_close(mlt_producer parent)3194 static void producer_close( mlt_producer parent )
3195 {
3196 // Remove this instance from the cache
3197 mlt_service_cache_purge( MLT_PRODUCER_SERVICE(parent) );
3198
3199 // Close the parent
3200 parent->close = NULL;
3201 mlt_producer_close( parent );
3202
3203 // Free the memory
3204 free( parent );
3205 }
3206