1 /*****************************************************************************
2 * encoder.c: video and audio encoder using the ffmpeg library
3 *****************************************************************************
4 * Copyright (C) 1999-2004 VideoLAN
5 * $Id: encoder.c,v 1.24 2004/03/03 11:29:26 massiot Exp $
7 * Authors: Laurent Aimar <fenrir@via.ecp.fr>
8 * Gildas Bazin <gbazin@netcourrier.com>
9 * Christophe Massiot <massiot@via.ecp.fr>
11 * This program is free software; you can redistribute it and/or modify
12 * it under the terms of the GNU General Public License as published by
13 * the Free Software Foundation; either version 2 of the License, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; if not, write to the Free Software
23 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
24 *****************************************************************************/
26 /*****************************************************************************
28 *****************************************************************************/
32 #include <vlc/decoder.h>
36 #ifdef HAVE_FFMPEG_AVCODEC_H
37 # include <ffmpeg/avcodec.h>
42 #if LIBAVCODEC_BUILD < 4704
43 # define AV_NOPTS_VALUE 0
48 #define AVCODEC_MAX_VIDEO_FRAME_SIZE (3*1024*1024)
49 #define HURRY_UP_GUARD1 (450000)
50 #define HURRY_UP_GUARD2 (300000)
51 #define HURRY_UP_GUARD3 (100000)
53 /*****************************************************************************
55 *****************************************************************************/
56 int E_(OpenEncoder) ( vlc_object_t * );
57 void E_(CloseEncoder)( vlc_object_t * );
59 static block_t *EncodeVideo( encoder_t *, picture_t * );
60 static block_t *EncodeAudio( encoder_t *, aout_buffer_t * );
62 struct thread_context_t;
63 static int FfmpegThread( struct thread_context_t *p_context );
64 static int FfmpegExecute( AVCodecContext *s,
65 int (*pf_func)(AVCodecContext *c2, void *arg2),
66 void **arg, int *ret, int count );
68 /*****************************************************************************
69 * thread_context_t : for multithreaded encoding
70 *****************************************************************************/
71 #if LIBAVCODEC_BUILD >= 4702
72 struct thread_context_t
76 AVCodecContext *p_context;
77 int (* pf_func)(AVCodecContext *c, void *arg);
83 vlc_bool_t b_work, b_done;
87 /*****************************************************************************
88 * encoder_sys_t : ffmpeg encoder descriptor
89 *****************************************************************************/
96 AVCodecContext *p_context;
107 mtime_t i_last_ref_pts;
108 mtime_t i_buggy_pts_detect;
120 /*****************************************************************************
121 * OpenEncoder: probe the encoder
122 *****************************************************************************/
123 extern int16_t ff_mpeg4_default_intra_matrix[];
124 extern int16_t ff_mpeg4_default_non_intra_matrix[];
126 int E_(OpenEncoder)( vlc_object_t *p_this )
128 encoder_t *p_enc = (encoder_t *)p_this;
129 encoder_sys_t *p_sys = p_enc->p_sys;
130 AVCodecContext *p_context;
132 int i_codec_id, i_cat;
135 if( !E_(GetFfmpegCodec)( p_enc->fmt_out.i_codec, &i_cat, &i_codec_id,
138 if( E_(GetFfmpegChroma)( p_enc->fmt_out.i_codec ) < 0 )
140 /* handed chroma output */
144 i_codec_id = CODEC_ID_RAWVIDEO;
145 psz_namecodec = "Raw video";
149 if( p_enc->fmt_out.i_cat == VIDEO_ES && i_cat != VIDEO_ES )
151 msg_Err( p_enc, "\"%s\" is not a video encoder", psz_namecodec );
155 if( p_enc->fmt_out.i_cat == AUDIO_ES && i_cat != AUDIO_ES )
157 msg_Err( p_enc, "\"%s\" is not an audio encoder", psz_namecodec );
161 /* Initialization must be done before avcodec_find_decoder() */
162 E_(InitLibavcodec)(p_this);
164 p_codec = avcodec_find_encoder( i_codec_id );
167 msg_Err( p_enc, "cannot find encoder %s", psz_namecodec );
171 /* Allocate the memory needed to store the decoder's structure */
172 if( ( p_sys = (encoder_sys_t *)malloc(sizeof(encoder_sys_t)) ) == NULL )
174 msg_Err( p_enc, "out of memory" );
177 p_enc->p_sys = p_sys;
178 p_sys->p_codec = p_codec;
180 p_enc->pf_encode_video = EncodeVideo;
181 p_enc->pf_encode_audio = EncodeAudio;
183 p_sys->p_buffer_out = NULL;
184 p_sys->p_buffer = NULL;
187 p_sys->p_context = p_context = avcodec_alloc_context();
189 /* Set CPU capabilities */
190 p_context->dsp_mask = 0;
191 if( !(p_enc->p_libvlc->i_cpu & CPU_CAPABILITY_MMX) )
193 p_context->dsp_mask |= FF_MM_MMX;
195 if( !(p_enc->p_libvlc->i_cpu & CPU_CAPABILITY_MMXEXT) )
197 p_context->dsp_mask |= FF_MM_MMXEXT;
199 if( !(p_enc->p_libvlc->i_cpu & CPU_CAPABILITY_3DNOW) )
201 p_context->dsp_mask |= FF_MM_3DNOW;
203 if( !(p_enc->p_libvlc->i_cpu & CPU_CAPABILITY_SSE) )
205 p_context->dsp_mask |= FF_MM_SSE;
206 p_context->dsp_mask |= FF_MM_SSE2;
209 /* Make sure we get extradata filled by the encoder */
210 p_context->extradata_size = 0;
211 p_context->extradata = NULL;
212 p_context->flags |= CODEC_FLAG_GLOBAL_HEADER;
214 if( p_enc->fmt_in.i_cat == VIDEO_ES )
216 if( !p_enc->fmt_in.video.i_width || !p_enc->fmt_in.video.i_height )
218 msg_Warn( p_enc, "invalid size %ix%i", p_enc->fmt_in.video.i_width,
219 p_enc->fmt_in.video.i_height );
224 p_context->width = p_enc->fmt_in.video.i_width;
225 p_context->height = p_enc->fmt_in.video.i_height;
227 p_context->frame_rate = p_enc->fmt_in.video.i_frame_rate;
228 p_context->frame_rate_base= p_enc->fmt_in.video.i_frame_rate_base;
230 /* Defaults from ffmpeg.c */
231 p_context->qblur = 0.5;
232 p_context->qcompress = 0.5;
233 p_context->b_quant_offset = 1.25;
234 p_context->b_quant_factor = 1.25;
235 p_context->i_quant_offset = 0.0;
236 p_context->i_quant_factor = -0.8;
238 p_context->gop_size = p_enc->i_key_int > 0 ? p_enc->i_key_int : 50;
239 p_context->max_b_frames =
240 __MIN( p_enc->i_b_frames, FF_MAX_B_FRAMES );
241 p_context->b_frame_strategy = 0;
243 #if LIBAVCODEC_BUILD >= 4687
244 p_context->sample_aspect_ratio =
245 (AVRational){ p_enc->fmt_in.video.i_aspect *
246 (int64_t)p_context->height / p_context->width,
247 VOUT_ASPECT_FACTOR };
249 p_context->aspect_ratio = ((float)p_enc->fmt_in.video.i_aspect) /
253 p_sys->p_buffer_out = malloc( AVCODEC_MAX_VIDEO_FRAME_SIZE );
255 p_enc->fmt_in.i_codec = VLC_FOURCC('I','4','2','0');
257 if ( p_enc->b_strict_rc )
259 p_context->rc_max_rate = p_enc->fmt_out.i_bitrate;
260 p_context->rc_buffer_size = p_enc->i_rc_buffer_size;
261 p_context->rc_buffer_aggressivity = p_enc->f_rc_buffer_aggressivity;
264 if ( p_enc->f_i_quant_factor != 0.0 )
266 p_context->i_quant_factor = p_enc->f_i_quant_factor;
269 #if LIBAVCODEC_BUILD >= 4690
270 p_context->noise_reduction = p_enc->i_noise_reduction;
273 if ( p_enc->b_mpeg4_matrix )
275 p_context->intra_matrix = ff_mpeg4_default_intra_matrix;
276 p_context->inter_matrix = ff_mpeg4_default_non_intra_matrix;
279 if ( p_enc->b_pre_me )
281 p_context->pre_me = 1;
282 p_context->me_pre_cmp = FF_CMP_CHROMA;
285 if ( p_enc->b_interlace )
287 p_context->flags |= CODEC_FLAG_INTERLACED_DCT;
288 #if LIBAVCODEC_BUILD >= 4698
289 p_context->flags |= CODEC_FLAG_INTERLACED_ME;
293 if ( p_enc->b_trellis )
295 p_context->flags |= CODEC_FLAG_TRELLIS_QUANT;
298 #if LIBAVCODEC_BUILD >= 4702
299 if ( p_enc->i_threads >= 1 )
301 p_context->thread_count = p_enc->i_threads;
305 if( p_enc->i_vtolerance > 0 )
307 p_context->bit_rate_tolerance = p_enc->i_vtolerance;
310 p_context->mb_qmin = p_context->qmin = p_enc->i_qmin;
311 p_context->mb_qmax = p_context->qmax = p_enc->i_qmax;
312 p_context->max_qdiff = 3;
314 p_context->mb_decision = p_enc->i_hq;
316 else if( p_enc->fmt_in.i_cat == AUDIO_ES )
318 p_enc->fmt_in.i_codec = AOUT_FMT_S16_NE;
319 p_context->sample_rate = p_enc->fmt_in.audio.i_rate;
320 p_context->channels = p_enc->fmt_in.audio.i_channels;
321 p_sys->i_frame_size = p_context->frame_size * 2 * p_context->channels;
322 p_sys->p_buffer = malloc( p_sys->i_frame_size );
323 p_sys->p_buffer_out = malloc( 2 * AVCODEC_MAX_AUDIO_FRAME_SIZE );
326 /* Misc parameters */
327 p_context->bit_rate = p_enc->fmt_out.i_bitrate;
329 if( i_codec_id == CODEC_ID_RAWVIDEO )
331 /* XXX: hack: Force same codec (will be handled by transcode) */
332 p_enc->fmt_in.i_codec = p_enc->fmt_out.i_codec;
333 p_context->pix_fmt = E_(GetFfmpegChroma)( p_enc->fmt_in.i_codec );
336 /* Make sure we get extradata filled by the encoder */
337 p_context->extradata_size = 0;
338 p_context->extradata = NULL;
339 p_context->flags |= CODEC_FLAG_GLOBAL_HEADER;
341 if( avcodec_open( p_context, p_codec ) )
343 if( p_enc->fmt_in.i_cat == AUDIO_ES && p_context->channels > 2 )
345 p_context->channels = 2;
346 p_enc->fmt_in.audio.i_channels = 2; // FIXME
347 if( avcodec_open( p_context, p_codec ) )
349 msg_Err( p_enc, "cannot open encoder" );
353 msg_Warn( p_enc, "stereo mode selected (codec limitation)" );
357 msg_Err( p_enc, "cannot open encoder" );
363 p_enc->fmt_out.i_extra = p_context->extradata_size;
364 p_enc->fmt_out.p_extra = p_context->extradata;
365 p_context->flags &= ~CODEC_FLAG_GLOBAL_HEADER;
367 if( p_enc->fmt_in.i_cat == AUDIO_ES )
369 p_sys->i_frame_size = p_context->frame_size * 2 * p_context->channels;
370 p_sys->p_buffer = malloc( p_sys->i_frame_size );
373 p_sys->i_last_ref_pts = 0;
374 p_sys->i_buggy_pts_detect = 0;
375 p_sys->i_samples_delay = 0;
377 p_sys->i_last_pts = 0;
379 msg_Dbg( p_enc, "found encoder %s", psz_namecodec );
384 /****************************************************************************
385 * Ffmpeg threading system
386 ****************************************************************************/
387 #if LIBAVCODEC_BUILD >= 4702
388 static int FfmpegThread( struct thread_context_t *p_context )
390 while ( !p_context->b_die && !p_context->b_error )
392 vlc_mutex_lock( &p_context->lock );
393 while ( !p_context->b_work && !p_context->b_die && !p_context->b_error )
395 vlc_cond_wait( &p_context->cond, &p_context->lock );
397 p_context->b_work = 0;
398 vlc_mutex_unlock( &p_context->lock );
399 if ( p_context->b_die || p_context->b_error )
402 if ( p_context->pf_func )
404 p_context->i_ret = p_context->pf_func( p_context->p_context,
408 vlc_mutex_lock( &p_context->lock );
409 p_context->b_done = 1;
410 vlc_cond_signal( &p_context->cond );
411 vlc_mutex_unlock( &p_context->lock );
417 static int FfmpegExecute( AVCodecContext *s,
418 int (*pf_func)(AVCodecContext *c2, void *arg2),
419 void **arg, int *ret, int count )
421 struct thread_context_t ** pp_contexts =
422 (struct thread_context_t **)s->thread_opaque;
425 /* Note, we can be certain that this is not called with the same
426 * AVCodecContext by different threads at the same time */
427 for ( i = 0; i < count; i++ )
429 vlc_mutex_lock( &pp_contexts[i]->lock );
430 pp_contexts[i]->arg = arg[i];
431 pp_contexts[i]->pf_func = pf_func;
432 pp_contexts[i]->i_ret = 12345;
433 pp_contexts[i]->b_work = 1;
434 vlc_cond_signal( &pp_contexts[i]->cond );
435 vlc_mutex_unlock( &pp_contexts[i]->lock );
437 for ( i = 0; i < count; i++ )
439 vlc_mutex_lock( &pp_contexts[i]->lock );
440 while ( !pp_contexts[i]->b_done )
442 vlc_cond_wait( &pp_contexts[i]->cond, &pp_contexts[i]->lock );
444 pp_contexts[i]->b_done = 0;
445 pp_contexts[i]->pf_func = NULL;
446 vlc_mutex_unlock( &pp_contexts[i]->lock );
450 ret[i] = pp_contexts[i]->i_ret;
458 /****************************************************************************
459 * EncodeVideo: the whole thing
460 ****************************************************************************/
461 static block_t *EncodeVideo( encoder_t *p_enc, picture_t *p_pict )
463 encoder_sys_t *p_sys = p_enc->p_sys;
467 #if LIBAVCODEC_BUILD >= 4702
468 if ( !p_sys->b_inited && p_enc->i_threads >= 1 )
470 struct thread_context_t ** pp_contexts;
474 pp_contexts = malloc( sizeof(struct thread_context_t *)
475 * p_enc->i_threads );
476 p_sys->p_context->thread_opaque = (void *)pp_contexts;
478 for ( i = 0; i < p_enc->i_threads; i++ )
480 pp_contexts[i] = vlc_object_create( p_enc,
481 sizeof(struct thread_context_t) );
482 pp_contexts[i]->p_context = p_sys->p_context;
483 vlc_mutex_init( p_enc, &pp_contexts[i]->lock );
484 vlc_cond_init( p_enc, &pp_contexts[i]->cond );
485 pp_contexts[i]->b_work = 0;
486 pp_contexts[i]->b_done = 0;
487 if ( vlc_thread_create( pp_contexts[i], "encoder", FfmpegThread,
488 VLC_THREAD_PRIORITY_VIDEO, VLC_FALSE ) )
490 msg_Err( p_enc, "cannot spawn encoder thread, expect to die soon" );
495 p_sys->p_context->execute = FfmpegExecute;
499 memset( &frame, 0, sizeof( AVFrame ) );
500 for( i_plane = 0; i_plane < p_pict->i_planes; i_plane++ )
502 frame.data[i_plane] = p_pict->p[i_plane].p_pixels;
503 frame.linesize[i_plane] = p_pict->p[i_plane].i_pitch;
506 /* Let ffmpeg select the frame type */
509 frame.repeat_pict = p_pict->i_nb_fields;
511 #if LIBAVCODEC_BUILD >= 4685
512 frame.interlaced_frame = !p_pict->b_progressive;
513 frame.top_field_first = p_pict->b_top_field_first;
516 /* Set the pts of the frame being encoded (segfaults with mpeg4!)*/
517 if( p_enc->fmt_out.i_codec == VLC_FOURCC( 'm', 'p', 'g', 'v' ) ||
518 p_enc->fmt_out.i_codec == VLC_FOURCC( 'm', 'p', '1', 'v' ) ||
519 p_enc->fmt_out.i_codec == VLC_FOURCC( 'm', 'p', '2', 'v' ) )
521 frame.pts = p_pict->date ? p_pict->date : AV_NOPTS_VALUE;
523 if ( p_enc->b_hurry_up && frame.pts != AV_NOPTS_VALUE )
525 mtime_t current_date = mdate();
527 if ( current_date + HURRY_UP_GUARD3 > frame.pts )
529 p_sys->p_context->mb_decision = FF_MB_DECISION_SIMPLE;
530 p_sys->p_context->flags &= ~CODEC_FLAG_TRELLIS_QUANT;
531 msg_Dbg( p_enc, "hurry up mode 3" );
535 p_sys->p_context->mb_decision = p_enc->i_hq;
537 if ( current_date + HURRY_UP_GUARD2 > frame.pts )
539 p_sys->p_context->flags &= ~CODEC_FLAG_TRELLIS_QUANT;
540 #if LIBAVCODEC_BUILD >= 4690
541 p_sys->p_context->noise_reduction = p_enc->i_noise_reduction
542 + (HURRY_UP_GUARD2 + current_date - frame.pts) / 500;
544 msg_Dbg( p_enc, "hurry up mode 2" );
548 if ( p_enc->b_trellis )
549 p_sys->p_context->flags |= CODEC_FLAG_TRELLIS_QUANT;
550 #if LIBAVCODEC_BUILD >= 4690
551 p_sys->p_context->noise_reduction =
552 p_enc->i_noise_reduction;
557 if ( current_date + HURRY_UP_GUARD1 > frame.pts )
559 frame.pict_type = FF_P_TYPE;
560 /* msg_Dbg( p_enc, "hurry up mode 1 %lld", current_date + HURRY_UP_GUARD1 - frame.pts ); */
566 frame.pts = AV_NOPTS_VALUE;
569 if ( frame.pts != AV_NOPTS_VALUE )
571 if ( p_sys->i_last_pts == frame.pts )
574 "almost fed libavcodec with two frames with the same PTS (" I64Fd ")",
580 p_sys->i_last_pts = frame.pts;
584 i_out = avcodec_encode_video( p_sys->p_context, p_sys->p_buffer_out,
585 AVCODEC_MAX_VIDEO_FRAME_SIZE, &frame );
589 block_t *p_block = block_New( p_enc, i_out );
590 memcpy( p_block->p_buffer, p_sys->p_buffer_out, i_out );
592 if( p_sys->p_context->coded_frame->pts != AV_NOPTS_VALUE &&
593 p_sys->i_buggy_pts_detect != p_sys->p_context->coded_frame->pts )
595 p_sys->i_buggy_pts_detect = p_sys->p_context->coded_frame->pts;
597 /* FIXME, 3-2 pulldown is not handled correctly */
598 p_block->i_length = I64C(1000000) *
599 p_enc->fmt_in.video.i_frame_rate_base /
600 p_enc->fmt_in.video.i_frame_rate;
601 p_block->i_pts = p_sys->p_context->coded_frame->pts;
603 if( !p_sys->p_context->delay ||
604 ( p_sys->p_context->coded_frame->pict_type != FF_I_TYPE &&
605 p_sys->p_context->coded_frame->pict_type != FF_P_TYPE ) )
607 p_block->i_dts = p_block->i_pts;
611 if( p_sys->i_last_ref_pts )
613 p_block->i_dts = p_sys->i_last_ref_pts;
617 /* Let's put something sensible */
618 p_block->i_dts = p_block->i_pts;
621 p_sys->i_last_ref_pts = p_block->i_pts;
626 /* Buggy libavcodec which doesn't update coded_frame->pts
628 p_block->i_length = I64C(1000000) *
629 p_enc->fmt_in.video.i_frame_rate_base /
630 p_enc->fmt_in.video.i_frame_rate;
631 p_block->i_dts = p_block->i_pts = p_pict->date;
634 switch ( p_sys->p_context->coded_frame->pict_type )
637 p_block->i_flags |= BLOCK_FLAG_TYPE_I;
640 p_block->i_flags |= BLOCK_FLAG_TYPE_P;
643 p_block->i_flags |= BLOCK_FLAG_TYPE_B;
653 /****************************************************************************
654 * EncodeAudio: the whole thing
655 ****************************************************************************/
656 static block_t *EncodeAudio( encoder_t *p_enc, aout_buffer_t *p_aout_buf )
658 encoder_sys_t *p_sys = p_enc->p_sys;
659 block_t *p_block, *p_chain = NULL;
661 char *p_buffer = p_aout_buf->p_buffer;
662 int i_samples = p_aout_buf->i_nb_samples;
663 int i_samples_delay = p_sys->i_samples_delay;
665 p_sys->i_pts = p_aout_buf->start_date -
666 (mtime_t)1000000 * (mtime_t)p_sys->i_samples_delay /
667 (mtime_t)p_enc->fmt_in.audio.i_rate;
669 p_sys->i_samples_delay += i_samples;
671 while( p_sys->i_samples_delay >= p_sys->p_context->frame_size )
676 if( i_samples_delay )
678 /* Take care of the left-over from last time */
679 int i_delay_size = i_samples_delay * 2 *
680 p_sys->p_context->channels;
681 int i_size = p_sys->i_frame_size - i_delay_size;
683 p_samples = (int16_t *)p_sys->p_buffer;
684 memcpy( p_sys->p_buffer + i_delay_size, p_buffer, i_size );
685 p_buffer -= i_delay_size;
686 i_samples += i_samples_delay;
691 p_samples = (int16_t *)p_buffer;
694 i_out = avcodec_encode_audio( p_sys->p_context, p_sys->p_buffer_out,
695 2 * AVCODEC_MAX_AUDIO_FRAME_SIZE,
699 msg_Warn( p_enc, "avcodec_encode_audio: %d", i_out );
701 if( i_out < 0 ) break;
703 p_buffer += p_sys->i_frame_size;
704 p_sys->i_samples_delay -= p_sys->p_context->frame_size;
705 i_samples -= p_sys->p_context->frame_size;
707 if( i_out == 0 ) continue;
709 p_block = block_New( p_enc, i_out );
710 memcpy( p_block->p_buffer, p_sys->p_buffer_out, i_out );
712 p_block->i_length = (mtime_t)1000000 *
713 (mtime_t)p_sys->p_context->frame_size /
714 (mtime_t)p_sys->p_context->sample_rate;
716 p_block->i_dts = p_block->i_pts = p_sys->i_pts;
719 p_sys->i_pts += p_block->i_length;
720 block_ChainAppend( &p_chain, p_block );
723 /* Backup the remaining raw samples */
726 memcpy( p_sys->p_buffer + i_samples_delay * 2 *
727 p_sys->p_context->channels, p_buffer,
728 i_samples * 2 * p_sys->p_context->channels );
734 /*****************************************************************************
735 * CloseEncoder: ffmpeg encoder destruction
736 *****************************************************************************/
737 void E_(CloseEncoder)( vlc_object_t *p_this )
739 encoder_t *p_enc = (encoder_t *)p_this;
740 encoder_sys_t *p_sys = p_enc->p_sys;
742 #if LIBAVCODEC_BUILD >= 4702
743 if ( p_sys->b_inited && p_enc->i_threads >= 1 )
746 struct thread_context_t ** pp_contexts =
747 (struct thread_context_t **)p_sys->p_context->thread_opaque;
748 for ( i = 0; i < p_enc->i_threads; i++ )
750 pp_contexts[i]->b_die = 1;
751 vlc_cond_signal( &pp_contexts[i]->cond );
752 vlc_thread_join( pp_contexts[i] );
753 vlc_mutex_destroy( &pp_contexts[i]->lock );
754 vlc_cond_destroy( &pp_contexts[i]->cond );
755 vlc_object_destroy( pp_contexts[i] );
762 avcodec_close( p_sys->p_context );
763 free( p_sys->p_context );
765 if( p_sys->p_buffer ) free( p_sys->p_buffer );
766 if( p_sys->p_buffer_out ) free( p_sys->p_buffer_out );