]> git.sesse.net Git - ffmpeg/blob - libswresample/swresample.c
lavf/http: deprecate user-agent option
[ffmpeg] / libswresample / swresample.c
1 /*
2  * Copyright (C) 2011-2013 Michael Niedermayer (michaelni@gmx.at)
3  *
4  * This file is part of libswresample
5  *
6  * libswresample is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU Lesser General Public
8  * License as published by the Free Software Foundation; either
9  * version 2.1 of the License, or (at your option) any later version.
10  *
11  * libswresample is distributed in the hope that it will be useful,
12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14  * Lesser General Public License for more details.
15  *
16  * You should have received a copy of the GNU Lesser General Public
17  * License along with libswresample; if not, write to the Free Software
18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19  */
20
21 #include "libavutil/opt.h"
22 #include "swresample_internal.h"
23 #include "audioconvert.h"
24 #include "libavutil/avassert.h"
25 #include "libavutil/channel_layout.h"
26 #include "libavutil/internal.h"
27
28 #include <float.h>
29
30 #define ALIGN 32
31
32 #include "libavutil/ffversion.h"
33 const char swr_ffversion[] = "FFmpeg version " FFMPEG_VERSION;
34
35 unsigned swresample_version(void)
36 {
37     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
38     return LIBSWRESAMPLE_VERSION_INT;
39 }
40
41 const char *swresample_configuration(void)
42 {
43     return FFMPEG_CONFIGURATION;
44 }
45
46 const char *swresample_license(void)
47 {
48 #define LICENSE_PREFIX "libswresample license: "
49     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
50 }
51
52 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
53     if(!s || s->in_convert) // s needs to be allocated but not initialized
54         return AVERROR(EINVAL);
55     s->channel_map = channel_map;
56     return 0;
57 }
58
59 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
60                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
61                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
62                                       int log_offset, void *log_ctx){
63     if(!s) s= swr_alloc();
64     if(!s) return NULL;
65
66     s->log_level_offset= log_offset;
67     s->log_ctx= log_ctx;
68
69     if (av_opt_set_int(s, "ocl", out_ch_layout,   0) < 0)
70         goto fail;
71
72     if (av_opt_set_int(s, "osf", out_sample_fmt,  0) < 0)
73         goto fail;
74
75     if (av_opt_set_int(s, "osr", out_sample_rate, 0) < 0)
76         goto fail;
77
78     if (av_opt_set_int(s, "icl", in_ch_layout,    0) < 0)
79         goto fail;
80
81     if (av_opt_set_int(s, "isf", in_sample_fmt,   0) < 0)
82         goto fail;
83
84     if (av_opt_set_int(s, "isr", in_sample_rate,  0) < 0)
85         goto fail;
86
87     if (av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0) < 0)
88         goto fail;
89
90     if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
91         goto fail;
92
93     if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
94         goto fail;
95
96     av_opt_set_int(s, "uch", 0, 0);
97     return s;
98 fail:
99     av_log(s, AV_LOG_ERROR, "Failed to set option\n");
100     swr_free(&s);
101     return NULL;
102 }
103
104 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
105     a->fmt   = fmt;
106     a->bps   = av_get_bytes_per_sample(fmt);
107     a->planar= av_sample_fmt_is_planar(fmt);
108     if (a->ch_count == 1)
109         a->planar = 1;
110 }
111
112 static void free_temp(AudioData *a){
113     av_free(a->data);
114     memset(a, 0, sizeof(*a));
115 }
116
117 static void clear_context(SwrContext *s){
118     s->in_buffer_index= 0;
119     s->in_buffer_count= 0;
120     s->resample_in_constraint= 0;
121     memset(s->in.ch, 0, sizeof(s->in.ch));
122     memset(s->out.ch, 0, sizeof(s->out.ch));
123     free_temp(&s->postin);
124     free_temp(&s->midbuf);
125     free_temp(&s->preout);
126     free_temp(&s->in_buffer);
127     free_temp(&s->silence);
128     free_temp(&s->drop_temp);
129     free_temp(&s->dither.noise);
130     free_temp(&s->dither.temp);
131     swri_audio_convert_free(&s-> in_convert);
132     swri_audio_convert_free(&s->out_convert);
133     swri_audio_convert_free(&s->full_convert);
134     swri_rematrix_free(s);
135
136     s->delayed_samples_fixup = 0;
137     s->flushed = 0;
138 }
139
140 av_cold void swr_free(SwrContext **ss){
141     SwrContext *s= *ss;
142     if(s){
143         clear_context(s);
144         if (s->resampler)
145             s->resampler->free(&s->resample);
146     }
147
148     av_freep(ss);
149 }
150
151 av_cold void swr_close(SwrContext *s){
152     clear_context(s);
153 }
154
155 av_cold int swr_init(struct SwrContext *s){
156     int ret;
157     char l1[1024], l2[1024];
158
159     clear_context(s);
160
161     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
162         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
163         return AVERROR(EINVAL);
164     }
165     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
166         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
167         return AVERROR(EINVAL);
168     }
169
170     s->out.ch_count  = s-> user_out_ch_count;
171     s-> in.ch_count  = s->  user_in_ch_count;
172     s->used_ch_count = s->user_used_ch_count;
173
174     s-> in_ch_layout = s-> user_in_ch_layout;
175     s->out_ch_layout = s->user_out_ch_layout;
176
177     s->int_sample_fmt= s->user_int_sample_fmt;
178
179     s->dither.method = s->user_dither_method;
180
181     if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
182         av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
183         s->in_ch_layout = 0;
184     }
185
186     if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
187         av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
188         s->out_ch_layout = 0;
189     }
190
191     switch(s->engine){
192 #if CONFIG_LIBSOXR
193         case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
194 #endif
195         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
196         default:
197             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
198             return AVERROR(EINVAL);
199     }
200
201     if(!s->used_ch_count)
202         s->used_ch_count= s->in.ch_count;
203
204     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
205         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
206         s-> in_ch_layout= 0;
207     }
208
209     if(!s-> in_ch_layout)
210         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
211     if(!s->out_ch_layout)
212         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
213
214     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
215                  s->rematrix_custom;
216
217     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
218         if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
219            && av_get_bytes_per_sample(s->out_sample_fmt) <= 2){
220             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
221         }else if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
222            && !s->rematrix
223            && s->out_sample_rate==s->in_sample_rate
224            && !(s->flags & SWR_FLAG_RESAMPLE)){
225             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
226         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
227                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
228                  && !s->rematrix
229                  && s->engine != SWR_ENGINE_SOXR){
230             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
231         }else if(av_get_bytes_per_sample(s->in_sample_fmt) <= 4){
232             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
233         }else{
234             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
235         }
236     }
237     av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
238
239     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
240         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
241         &&s->int_sample_fmt != AV_SAMPLE_FMT_S64P
242         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
243         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
244         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, S16/S32/S64/FLT/DBL is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
245         return AVERROR(EINVAL);
246     }
247
248     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
249     set_audiodata_fmt(&s->out, s->out_sample_fmt);
250
251     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
252         if (!s->async && s->min_compensation >= FLT_MAX/2)
253             s->async = 1;
254         s->firstpts =
255         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
256     } else
257         s->firstpts = AV_NOPTS_VALUE;
258
259     if (s->async) {
260         if (s->min_compensation >= FLT_MAX/2)
261             s->min_compensation = 0.001;
262         if (s->async > 1.0001) {
263             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
264         }
265     }
266
267     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
268         s->resample = s->resampler->init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta, s->precision, s->cheby, s->exact_rational);
269         if (!s->resample) {
270             av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
271             return AVERROR(ENOMEM);
272         }
273     }else
274         s->resampler->free(&s->resample);
275     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
276         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
277         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
278         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
279         && s->resample){
280         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
281         ret = AVERROR(EINVAL);
282         goto fail;
283     }
284
285 #define RSC 1 //FIXME finetune
286     if(!s-> in.ch_count)
287         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
288     if(!s->used_ch_count)
289         s->used_ch_count= s->in.ch_count;
290     if(!s->out.ch_count)
291         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
292
293     if(!s-> in.ch_count){
294         av_assert0(!s->in_ch_layout);
295         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
296         ret = AVERROR(EINVAL);
297         goto fail;
298     }
299
300     av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
301     av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
302     if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
303         av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
304         ret = AVERROR(EINVAL);
305         goto fail;
306     }
307     if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
308         av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
309         ret = AVERROR(EINVAL);
310         goto fail;
311     }
312
313     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
314         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
315                "but there is not enough information to do it\n", l1, l2);
316         ret = AVERROR(EINVAL);
317         goto fail;
318     }
319
320 av_assert0(s->used_ch_count);
321 av_assert0(s->out.ch_count);
322     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
323
324     s->in_buffer= s->in;
325     s->silence  = s->in;
326     s->drop_temp= s->out;
327
328     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
329         goto fail;
330
331     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
332         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
333                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
334         return 0;
335     }
336
337     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
338                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
339     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
340                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
341
342     if (!s->in_convert || !s->out_convert) {
343         ret = AVERROR(ENOMEM);
344         goto fail;
345     }
346
347     s->postin= s->in;
348     s->preout= s->out;
349     s->midbuf= s->in;
350
351     if(s->channel_map){
352         s->postin.ch_count=
353         s->midbuf.ch_count= s->used_ch_count;
354         if(s->resample)
355             s->in_buffer.ch_count= s->used_ch_count;
356     }
357     if(!s->resample_first){
358         s->midbuf.ch_count= s->out.ch_count;
359         if(s->resample)
360             s->in_buffer.ch_count = s->out.ch_count;
361     }
362
363     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
364     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
365     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
366
367     if(s->resample){
368         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
369     }
370
371     av_assert0(!s->preout.count);
372     s->dither.noise = s->preout;
373     s->dither.temp  = s->preout;
374     if (s->dither.method > SWR_DITHER_NS) {
375         s->dither.noise.bps = 4;
376         s->dither.noise.fmt = AV_SAMPLE_FMT_FLTP;
377         s->dither.noise_scale = 1;
378     }
379
380     if(s->rematrix || s->dither.method) {
381         ret = swri_rematrix_init(s);
382         if (ret < 0)
383             goto fail;
384     }
385
386     return 0;
387 fail:
388     swr_close(s);
389     return ret;
390
391 }
392
393 int swri_realloc_audio(AudioData *a, int count){
394     int i, countb;
395     AudioData old;
396
397     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
398         return AVERROR(EINVAL);
399
400     if(a->count >= count)
401         return 0;
402
403     count*=2;
404
405     countb= FFALIGN(count*a->bps, ALIGN);
406     old= *a;
407
408     av_assert0(a->bps);
409     av_assert0(a->ch_count);
410
411     a->data= av_mallocz_array(countb, a->ch_count);
412     if(!a->data)
413         return AVERROR(ENOMEM);
414     for(i=0; i<a->ch_count; i++){
415         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
416         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
417     }
418     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
419     av_freep(&old.data);
420     a->count= count;
421
422     return 1;
423 }
424
425 static void copy(AudioData *out, AudioData *in,
426                  int count){
427     av_assert0(out->planar == in->planar);
428     av_assert0(out->bps == in->bps);
429     av_assert0(out->ch_count == in->ch_count);
430     if(out->planar){
431         int ch;
432         for(ch=0; ch<out->ch_count; ch++)
433             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
434     }else
435         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
436 }
437
438 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
439     int i;
440     if(!in_arg){
441         memset(out->ch, 0, sizeof(out->ch));
442     }else if(out->planar){
443         for(i=0; i<out->ch_count; i++)
444             out->ch[i]= in_arg[i];
445     }else{
446         for(i=0; i<out->ch_count; i++)
447             out->ch[i]= in_arg[0] + i*out->bps;
448     }
449 }
450
451 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
452     int i;
453     if(out->planar){
454         for(i=0; i<out->ch_count; i++)
455             in_arg[i]= out->ch[i];
456     }else{
457         in_arg[0]= out->ch[0];
458     }
459 }
460
461 /**
462  *
463  * out may be equal in.
464  */
465 static void buf_set(AudioData *out, AudioData *in, int count){
466     int ch;
467     if(in->planar){
468         for(ch=0; ch<out->ch_count; ch++)
469             out->ch[ch]= in->ch[ch] + count*out->bps;
470     }else{
471         for(ch=out->ch_count-1; ch>=0; ch--)
472             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
473     }
474 }
475
476 /**
477  *
478  * @return number of samples output per channel
479  */
480 static int resample(SwrContext *s, AudioData *out_param, int out_count,
481                              const AudioData * in_param, int in_count){
482     AudioData in, out, tmp;
483     int ret_sum=0;
484     int border=0;
485     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
486
487     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
488     av_assert1(s->in_buffer.planar   == in_param->planar);
489     av_assert1(s->in_buffer.fmt      == in_param->fmt);
490
491     tmp=out=*out_param;
492     in =  *in_param;
493
494     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
495                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
496     if (border == INT_MAX) {
497         return 0;
498     } else if (border < 0) {
499         return border;
500     } else if (border) {
501         buf_set(&in, &in, border);
502         in_count -= border;
503         s->resample_in_constraint = 0;
504     }
505
506     do{
507         int ret, size, consumed;
508         if(!s->resample_in_constraint && s->in_buffer_count){
509             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
510             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
511             out_count -= ret;
512             ret_sum += ret;
513             buf_set(&out, &out, ret);
514             s->in_buffer_count -= consumed;
515             s->in_buffer_index += consumed;
516
517             if(!in_count)
518                 break;
519             if(s->in_buffer_count <= border){
520                 buf_set(&in, &in, -s->in_buffer_count);
521                 in_count += s->in_buffer_count;
522                 s->in_buffer_count=0;
523                 s->in_buffer_index=0;
524                 border = 0;
525             }
526         }
527
528         if((s->flushed || in_count > padless) && !s->in_buffer_count){
529             s->in_buffer_index=0;
530             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
531             out_count -= ret;
532             ret_sum += ret;
533             buf_set(&out, &out, ret);
534             in_count -= consumed;
535             buf_set(&in, &in, consumed);
536         }
537
538         //TODO is this check sane considering the advanced copy avoidance below
539         size= s->in_buffer_index + s->in_buffer_count + in_count;
540         if(   size > s->in_buffer.count
541            && s->in_buffer_count + in_count <= s->in_buffer_index){
542             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
543             copy(&s->in_buffer, &tmp, s->in_buffer_count);
544             s->in_buffer_index=0;
545         }else
546             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
547                 return ret;
548
549         if(in_count){
550             int count= in_count;
551             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
552
553             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
554             copy(&tmp, &in, /*in_*/count);
555             s->in_buffer_count += count;
556             in_count -= count;
557             border += count;
558             buf_set(&in, &in, count);
559             s->resample_in_constraint= 0;
560             if(s->in_buffer_count != count || in_count)
561                 continue;
562             if (padless) {
563                 padless = 0;
564                 continue;
565             }
566         }
567         break;
568     }while(1);
569
570     s->resample_in_constraint= !!out_count;
571
572     return ret_sum;
573 }
574
575 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
576                                                       AudioData *in , int  in_count){
577     AudioData *postin, *midbuf, *preout;
578     int ret/*, in_max*/;
579     AudioData preout_tmp, midbuf_tmp;
580
581     if(s->full_convert){
582         av_assert0(!s->resample);
583         swri_audio_convert(s->full_convert, out, in, in_count);
584         return out_count;
585     }
586
587 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
588 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
589
590     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
591         return ret;
592     if(s->resample_first){
593         av_assert0(s->midbuf.ch_count == s->used_ch_count);
594         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
595             return ret;
596     }else{
597         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
598         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
599             return ret;
600     }
601     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
602         return ret;
603
604     postin= &s->postin;
605
606     midbuf_tmp= s->midbuf;
607     midbuf= &midbuf_tmp;
608     preout_tmp= s->preout;
609     preout= &preout_tmp;
610
611     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
612         postin= in;
613
614     if(s->resample_first ? !s->resample : !s->rematrix)
615         midbuf= postin;
616
617     if(s->resample_first ? !s->rematrix : !s->resample)
618         preout= midbuf;
619
620     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
621        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
622         if(preout==in){
623             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
624             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
625             copy(out, in, out_count);
626             return out_count;
627         }
628         else if(preout==postin) preout= midbuf= postin= out;
629         else if(preout==midbuf) preout= midbuf= out;
630         else                    preout= out;
631     }
632
633     if(in != postin){
634         swri_audio_convert(s->in_convert, postin, in, in_count);
635     }
636
637     if(s->resample_first){
638         if(postin != midbuf)
639             out_count= resample(s, midbuf, out_count, postin, in_count);
640         if(midbuf != preout)
641             swri_rematrix(s, preout, midbuf, out_count, preout==out);
642     }else{
643         if(postin != midbuf)
644             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
645         if(midbuf != preout)
646             out_count= resample(s, preout, out_count, midbuf, in_count);
647     }
648
649     if(preout != out && out_count){
650         AudioData *conv_src = preout;
651         if(s->dither.method){
652             int ch;
653             int dither_count= FFMAX(out_count, 1<<16);
654
655             if (preout == in) {
656                 conv_src = &s->dither.temp;
657                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
658                     return ret;
659             }
660
661             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
662                 return ret;
663             if(ret)
664                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
665                     if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
666                         return ret;
667             av_assert0(s->dither.noise.ch_count == preout->ch_count);
668
669             if(s->dither.noise_pos + out_count > s->dither.noise.count)
670                 s->dither.noise_pos = 0;
671
672             if (s->dither.method < SWR_DITHER_NS){
673                 if (s->mix_2_1_simd) {
674                     int len1= out_count&~15;
675                     int off = len1 * preout->bps;
676
677                     if(len1)
678                         for(ch=0; ch<preout->ch_count; ch++)
679                             s->mix_2_1_simd(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_simd_one, 0, 0, len1);
680                     if(out_count != len1)
681                         for(ch=0; ch<preout->ch_count; ch++)
682                             s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off + len1, s->native_one, 0, 0, out_count - len1);
683                 } else {
684                     for(ch=0; ch<preout->ch_count; ch++)
685                         s->mix_2_1_f(conv_src->ch[ch], preout->ch[ch], s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos, s->native_one, 0, 0, out_count);
686                 }
687             } else {
688                 switch(s->int_sample_fmt) {
689                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
690                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
691                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
692                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
693                 }
694             }
695             s->dither.noise_pos += out_count;
696         }
697 //FIXME packed doesn't need more than 1 chan here!
698         swri_audio_convert(s->out_convert, out, conv_src, out_count);
699     }
700     return out_count;
701 }
702
703 int swr_is_initialized(struct SwrContext *s) {
704     return !!s->in_buffer.ch_count;
705 }
706
707 int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
708                                                     const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
709     AudioData * in= &s->in;
710     AudioData *out= &s->out;
711     int av_unused max_output;
712
713     if (!swr_is_initialized(s)) {
714         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
715         return AVERROR(EINVAL);
716     }
717 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
718     max_output = swr_get_out_samples(s, in_count);
719 #endif
720
721     while(s->drop_output > 0){
722         int ret;
723         uint8_t *tmp_arg[SWR_CH_MAX];
724 #define MAX_DROP_STEP 16384
725         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
726             return ret;
727
728         reversefill_audiodata(&s->drop_temp, tmp_arg);
729         s->drop_output *= -1; //FIXME find a less hackish solution
730         ret = swr_convert(s, tmp_arg, FFMIN(-s->drop_output, MAX_DROP_STEP), in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesn't matter
731         s->drop_output *= -1;
732         in_count = 0;
733         if(ret>0) {
734             s->drop_output -= ret;
735             if (!s->drop_output && !out_arg)
736                 return 0;
737             continue;
738         }
739
740         av_assert0(s->drop_output);
741         return 0;
742     }
743
744     if(!in_arg){
745         if(s->resample){
746             if (!s->flushed)
747                 s->resampler->flush(s);
748             s->resample_in_constraint = 0;
749             s->flushed = 1;
750         }else if(!s->in_buffer_count){
751             return 0;
752         }
753     }else
754         fill_audiodata(in ,  (void*)in_arg);
755
756     fill_audiodata(out, out_arg);
757
758     if(s->resample){
759         int ret = swr_convert_internal(s, out, out_count, in, in_count);
760         if(ret>0 && !s->drop_output)
761             s->outpts += ret * (int64_t)s->in_sample_rate;
762
763         av_assert2(max_output < 0 || ret < 0 || ret <= max_output);
764
765         return ret;
766     }else{
767         AudioData tmp= *in;
768         int ret2=0;
769         int ret, size;
770         size = FFMIN(out_count, s->in_buffer_count);
771         if(size){
772             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
773             ret= swr_convert_internal(s, out, size, &tmp, size);
774             if(ret<0)
775                 return ret;
776             ret2= ret;
777             s->in_buffer_count -= ret;
778             s->in_buffer_index += ret;
779             buf_set(out, out, ret);
780             out_count -= ret;
781             if(!s->in_buffer_count)
782                 s->in_buffer_index = 0;
783         }
784
785         if(in_count){
786             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
787
788             if(in_count > out_count) { //FIXME move after swr_convert_internal
789                 if(   size > s->in_buffer.count
790                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
791                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
792                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
793                     s->in_buffer_index=0;
794                 }else
795                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
796                         return ret;
797             }
798
799             if(out_count){
800                 size = FFMIN(in_count, out_count);
801                 ret= swr_convert_internal(s, out, size, in, size);
802                 if(ret<0)
803                     return ret;
804                 buf_set(in, in, ret);
805                 in_count -= ret;
806                 ret2 += ret;
807             }
808             if(in_count){
809                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
810                 copy(&tmp, in, in_count);
811                 s->in_buffer_count += in_count;
812             }
813         }
814         if(ret2>0 && !s->drop_output)
815             s->outpts += ret2 * (int64_t)s->in_sample_rate;
816         av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
817         return ret2;
818     }
819 }
820
821 int swr_drop_output(struct SwrContext *s, int count){
822     const uint8_t *tmp_arg[SWR_CH_MAX];
823     s->drop_output += count;
824
825     if(s->drop_output <= 0)
826         return 0;
827
828     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
829     return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
830 }
831
832 int swr_inject_silence(struct SwrContext *s, int count){
833     int ret, i;
834     uint8_t *tmp_arg[SWR_CH_MAX];
835
836     if(count <= 0)
837         return 0;
838
839 #define MAX_SILENCE_STEP 16384
840     while (count > MAX_SILENCE_STEP) {
841         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
842             return ret;
843         count -= MAX_SILENCE_STEP;
844     }
845
846     if((ret=swri_realloc_audio(&s->silence, count))<0)
847         return ret;
848
849     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
850         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
851     } else
852         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
853
854     reversefill_audiodata(&s->silence, tmp_arg);
855     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
856     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
857     return ret;
858 }
859
860 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
861     if (s->resampler && s->resample){
862         return s->resampler->get_delay(s, base);
863     }else{
864         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
865     }
866 }
867
868 int swr_get_out_samples(struct SwrContext *s, int in_samples)
869 {
870     int64_t out_samples;
871
872     if (in_samples < 0)
873         return AVERROR(EINVAL);
874
875     if (s->resampler && s->resample) {
876         if (!s->resampler->get_out_samples)
877             return AVERROR(ENOSYS);
878         out_samples = s->resampler->get_out_samples(s, in_samples);
879     } else {
880         out_samples = s->in_buffer_count + in_samples;
881         av_assert0(s->out_sample_rate == s->in_sample_rate);
882     }
883
884     if (out_samples > INT_MAX)
885         return AVERROR(EINVAL);
886
887     return out_samples;
888 }
889
890 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
891     int ret;
892
893     if (!s || compensation_distance < 0)
894         return AVERROR(EINVAL);
895     if (!compensation_distance && sample_delta)
896         return AVERROR(EINVAL);
897     if (!s->resample) {
898         s->flags |= SWR_FLAG_RESAMPLE;
899         ret = swr_init(s);
900         if (ret < 0)
901             return ret;
902     }
903     if (!s->resampler->set_compensation){
904         return AVERROR(EINVAL);
905     }else{
906         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
907     }
908 }
909
910 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
911     if(pts == INT64_MIN)
912         return s->outpts;
913
914     if (s->firstpts == AV_NOPTS_VALUE)
915         s->outpts = s->firstpts = pts;
916
917     if(s->min_compensation >= FLT_MAX) {
918         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
919     } else {
920         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts + s->drop_output*(int64_t)s->in_sample_rate;
921         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
922
923         if(fabs(fdelta) > s->min_compensation) {
924             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
925                 int ret;
926                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
927                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
928                 if(ret<0){
929                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
930                 }
931             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
932                 int duration = s->out_sample_rate * s->soft_compensation_duration;
933                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
934                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
935                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
936                 swr_set_compensation(s, comp, duration);
937             }
938         }
939
940         return s->outpts;
941     }
942 }