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