lavf/bit: Use pkt->size instead of a constant for G.729 frame size.
[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->out_sample_rate == s->in_sample_rate
227                  && !(s->flags & SWR_FLAG_RESAMPLE)
228                  && s->engine != SWR_ENGINE_SOXR){
229             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
230         }else if(av_get_bytes_per_sample(s->in_sample_fmt) <= 4){
231             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
232         }else{
233             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
234         }
235     }
236     av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
237
238     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
239         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
240         &&s->int_sample_fmt != AV_SAMPLE_FMT_S64P
241         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
242         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
243         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));
244         return AVERROR(EINVAL);
245     }
246
247     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
248     set_audiodata_fmt(&s->out, s->out_sample_fmt);
249
250     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
251         if (!s->async && s->min_compensation >= FLT_MAX/2)
252             s->async = 1;
253         s->firstpts =
254         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
255     } else
256         s->firstpts = AV_NOPTS_VALUE;
257
258     if (s->async) {
259         if (s->min_compensation >= FLT_MAX/2)
260             s->min_compensation = 0.001;
261         if (s->async > 1.0001) {
262             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
263         }
264     }
265
266     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
267         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);
268         if (!s->resample) {
269             av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
270             return AVERROR(ENOMEM);
271         }
272     }else
273         s->resampler->free(&s->resample);
274     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
275         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
276         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
277         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
278         && s->resample){
279         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
280         ret = AVERROR(EINVAL);
281         goto fail;
282     }
283
284 #define RSC 1 //FIXME finetune
285     if(!s-> in.ch_count)
286         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
287     if(!s->used_ch_count)
288         s->used_ch_count= s->in.ch_count;
289     if(!s->out.ch_count)
290         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
291
292     if(!s-> in.ch_count){
293         av_assert0(!s->in_ch_layout);
294         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
295         ret = AVERROR(EINVAL);
296         goto fail;
297     }
298
299     av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
300     av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
301     if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
302         av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
303         ret = AVERROR(EINVAL);
304         goto fail;
305     }
306     if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
307         av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
308         ret = AVERROR(EINVAL);
309         goto fail;
310     }
311
312     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
313         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
314                "but there is not enough information to do it\n", l1, l2);
315         ret = AVERROR(EINVAL);
316         goto fail;
317     }
318
319 av_assert0(s->used_ch_count);
320 av_assert0(s->out.ch_count);
321     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
322
323     s->in_buffer= s->in;
324     s->silence  = s->in;
325     s->drop_temp= s->out;
326
327     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
328         goto fail;
329
330     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
331         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
332                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
333         return 0;
334     }
335
336     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
337                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
338     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
339                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
340
341     if (!s->in_convert || !s->out_convert) {
342         ret = AVERROR(ENOMEM);
343         goto fail;
344     }
345
346     s->postin= s->in;
347     s->preout= s->out;
348     s->midbuf= s->in;
349
350     if(s->channel_map){
351         s->postin.ch_count=
352         s->midbuf.ch_count= s->used_ch_count;
353         if(s->resample)
354             s->in_buffer.ch_count= s->used_ch_count;
355     }
356     if(!s->resample_first){
357         s->midbuf.ch_count= s->out.ch_count;
358         if(s->resample)
359             s->in_buffer.ch_count = s->out.ch_count;
360     }
361
362     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
363     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
364     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
365
366     if(s->resample){
367         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
368     }
369
370     av_assert0(!s->preout.count);
371     s->dither.noise = s->preout;
372     s->dither.temp  = s->preout;
373     if (s->dither.method > SWR_DITHER_NS) {
374         s->dither.noise.bps = 4;
375         s->dither.noise.fmt = AV_SAMPLE_FMT_FLTP;
376         s->dither.noise_scale = 1;
377     }
378
379     if(s->rematrix || s->dither.method) {
380         ret = swri_rematrix_init(s);
381         if (ret < 0)
382             goto fail;
383     }
384
385     return 0;
386 fail:
387     swr_close(s);
388     return ret;
389
390 }
391
392 int swri_realloc_audio(AudioData *a, int count){
393     int i, countb;
394     AudioData old;
395
396     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
397         return AVERROR(EINVAL);
398
399     if(a->count >= count)
400         return 0;
401
402     count*=2;
403
404     countb= FFALIGN(count*a->bps, ALIGN);
405     old= *a;
406
407     av_assert0(a->bps);
408     av_assert0(a->ch_count);
409
410     a->data= av_mallocz_array(countb, a->ch_count);
411     if(!a->data)
412         return AVERROR(ENOMEM);
413     for(i=0; i<a->ch_count; i++){
414         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
415         if(a->count && a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
416     }
417     if(a->count && !a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
418     av_freep(&old.data);
419     a->count= count;
420
421     return 1;
422 }
423
424 static void copy(AudioData *out, AudioData *in,
425                  int count){
426     av_assert0(out->planar == in->planar);
427     av_assert0(out->bps == in->bps);
428     av_assert0(out->ch_count == in->ch_count);
429     if(out->planar){
430         int ch;
431         for(ch=0; ch<out->ch_count; ch++)
432             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
433     }else
434         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
435 }
436
437 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
438     int i;
439     if(!in_arg){
440         memset(out->ch, 0, sizeof(out->ch));
441     }else if(out->planar){
442         for(i=0; i<out->ch_count; i++)
443             out->ch[i]= in_arg[i];
444     }else{
445         for(i=0; i<out->ch_count; i++)
446             out->ch[i]= in_arg[0] + i*out->bps;
447     }
448 }
449
450 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
451     int i;
452     if(out->planar){
453         for(i=0; i<out->ch_count; i++)
454             in_arg[i]= out->ch[i];
455     }else{
456         in_arg[0]= out->ch[0];
457     }
458 }
459
460 /**
461  *
462  * out may be equal in.
463  */
464 static void buf_set(AudioData *out, AudioData *in, int count){
465     int ch;
466     if(in->planar){
467         for(ch=0; ch<out->ch_count; ch++)
468             out->ch[ch]= in->ch[ch] + count*out->bps;
469     }else{
470         for(ch=out->ch_count-1; ch>=0; ch--)
471             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
472     }
473 }
474
475 /**
476  *
477  * @return number of samples output per channel
478  */
479 static int resample(SwrContext *s, AudioData *out_param, int out_count,
480                              const AudioData * in_param, int in_count){
481     AudioData in, out, tmp;
482     int ret_sum=0;
483     int border=0;
484     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
485
486     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
487     av_assert1(s->in_buffer.planar   == in_param->planar);
488     av_assert1(s->in_buffer.fmt      == in_param->fmt);
489
490     tmp=out=*out_param;
491     in =  *in_param;
492
493     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
494                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
495     if (border == INT_MAX) {
496         return 0;
497     } else if (border < 0) {
498         return border;
499     } else if (border) {
500         buf_set(&in, &in, border);
501         in_count -= border;
502         s->resample_in_constraint = 0;
503     }
504
505     do{
506         int ret, size, consumed;
507         if(!s->resample_in_constraint && s->in_buffer_count){
508             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
509             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
510             out_count -= ret;
511             ret_sum += ret;
512             buf_set(&out, &out, ret);
513             s->in_buffer_count -= consumed;
514             s->in_buffer_index += consumed;
515
516             if(!in_count)
517                 break;
518             if(s->in_buffer_count <= border){
519                 buf_set(&in, &in, -s->in_buffer_count);
520                 in_count += s->in_buffer_count;
521                 s->in_buffer_count=0;
522                 s->in_buffer_index=0;
523                 border = 0;
524             }
525         }
526
527         if((s->flushed || in_count > padless) && !s->in_buffer_count){
528             s->in_buffer_index=0;
529             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
530             out_count -= ret;
531             ret_sum += ret;
532             buf_set(&out, &out, ret);
533             in_count -= consumed;
534             buf_set(&in, &in, consumed);
535         }
536
537         //TODO is this check sane considering the advanced copy avoidance below
538         size= s->in_buffer_index + s->in_buffer_count + in_count;
539         if(   size > s->in_buffer.count
540            && s->in_buffer_count + in_count <= s->in_buffer_index){
541             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
542             copy(&s->in_buffer, &tmp, s->in_buffer_count);
543             s->in_buffer_index=0;
544         }else
545             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
546                 return ret;
547
548         if(in_count){
549             int count= in_count;
550             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
551
552             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
553             copy(&tmp, &in, /*in_*/count);
554             s->in_buffer_count += count;
555             in_count -= count;
556             border += count;
557             buf_set(&in, &in, count);
558             s->resample_in_constraint= 0;
559             if(s->in_buffer_count != count || in_count)
560                 continue;
561             if (padless) {
562                 padless = 0;
563                 continue;
564             }
565         }
566         break;
567     }while(1);
568
569     s->resample_in_constraint= !!out_count;
570
571     return ret_sum;
572 }
573
574 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
575                                                       AudioData *in , int  in_count){
576     AudioData *postin, *midbuf, *preout;
577     int ret/*, in_max*/;
578     AudioData preout_tmp, midbuf_tmp;
579
580     if(s->full_convert){
581         av_assert0(!s->resample);
582         swri_audio_convert(s->full_convert, out, in, in_count);
583         return out_count;
584     }
585
586 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
587 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
588
589     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
590         return ret;
591     if(s->resample_first){
592         av_assert0(s->midbuf.ch_count == s->used_ch_count);
593         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
594             return ret;
595     }else{
596         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
597         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
598             return ret;
599     }
600     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
601         return ret;
602
603     postin= &s->postin;
604
605     midbuf_tmp= s->midbuf;
606     midbuf= &midbuf_tmp;
607     preout_tmp= s->preout;
608     preout= &preout_tmp;
609
610     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
611         postin= in;
612
613     if(s->resample_first ? !s->resample : !s->rematrix)
614         midbuf= postin;
615
616     if(s->resample_first ? !s->rematrix : !s->resample)
617         preout= midbuf;
618
619     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
620        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
621         if(preout==in){
622             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
623             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
624             copy(out, in, out_count);
625             return out_count;
626         }
627         else if(preout==postin) preout= midbuf= postin= out;
628         else if(preout==midbuf) preout= midbuf= out;
629         else                    preout= out;
630     }
631
632     if(in != postin){
633         swri_audio_convert(s->in_convert, postin, in, in_count);
634     }
635
636     if(s->resample_first){
637         if(postin != midbuf)
638             out_count= resample(s, midbuf, out_count, postin, in_count);
639         if(midbuf != preout)
640             swri_rematrix(s, preout, midbuf, out_count, preout==out);
641     }else{
642         if(postin != midbuf)
643             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
644         if(midbuf != preout)
645             out_count= resample(s, preout, out_count, midbuf, in_count);
646     }
647
648     if(preout != out && out_count){
649         AudioData *conv_src = preout;
650         if(s->dither.method){
651             int ch;
652             int dither_count= FFMAX(out_count, 1<<16);
653
654             if (preout == in) {
655                 conv_src = &s->dither.temp;
656                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
657                     return ret;
658             }
659
660             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
661                 return ret;
662             if(ret)
663                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
664                     if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
665                         return ret;
666             av_assert0(s->dither.noise.ch_count == preout->ch_count);
667
668             if(s->dither.noise_pos + out_count > s->dither.noise.count)
669                 s->dither.noise_pos = 0;
670
671             if (s->dither.method < SWR_DITHER_NS){
672                 if (s->mix_2_1_simd) {
673                     int len1= out_count&~15;
674                     int off = len1 * preout->bps;
675
676                     if(len1)
677                         for(ch=0; ch<preout->ch_count; ch++)
678                             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);
679                     if(out_count != len1)
680                         for(ch=0; ch<preout->ch_count; ch++)
681                             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);
682                 } else {
683                     for(ch=0; ch<preout->ch_count; ch++)
684                         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);
685                 }
686             } else {
687                 switch(s->int_sample_fmt) {
688                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
689                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
690                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
691                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
692                 }
693             }
694             s->dither.noise_pos += out_count;
695         }
696 //FIXME packed doesn't need more than 1 chan here!
697         swri_audio_convert(s->out_convert, out, conv_src, out_count);
698     }
699     return out_count;
700 }
701
702 int swr_is_initialized(struct SwrContext *s) {
703     return !!s->in_buffer.ch_count;
704 }
705
706 int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
707                                                     const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
708     AudioData * in= &s->in;
709     AudioData *out= &s->out;
710     int av_unused max_output;
711
712     if (!swr_is_initialized(s)) {
713         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
714         return AVERROR(EINVAL);
715     }
716 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
717     max_output = swr_get_out_samples(s, in_count);
718 #endif
719
720     while(s->drop_output > 0){
721         int ret;
722         uint8_t *tmp_arg[SWR_CH_MAX];
723 #define MAX_DROP_STEP 16384
724         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
725             return ret;
726
727         reversefill_audiodata(&s->drop_temp, tmp_arg);
728         s->drop_output *= -1; //FIXME find a less hackish solution
729         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
730         s->drop_output *= -1;
731         in_count = 0;
732         if(ret>0) {
733             s->drop_output -= ret;
734             if (!s->drop_output && !out_arg)
735                 return 0;
736             continue;
737         }
738
739         av_assert0(s->drop_output);
740         return 0;
741     }
742
743     if(!in_arg){
744         if(s->resample){
745             if (!s->flushed)
746                 s->resampler->flush(s);
747             s->resample_in_constraint = 0;
748             s->flushed = 1;
749         }else if(!s->in_buffer_count){
750             return 0;
751         }
752     }else
753         fill_audiodata(in ,  (void*)in_arg);
754
755     fill_audiodata(out, out_arg);
756
757     if(s->resample){
758         int ret = swr_convert_internal(s, out, out_count, in, in_count);
759         if(ret>0 && !s->drop_output)
760             s->outpts += ret * (int64_t)s->in_sample_rate;
761
762         av_assert2(max_output < 0 || ret < 0 || ret <= max_output);
763
764         return ret;
765     }else{
766         AudioData tmp= *in;
767         int ret2=0;
768         int ret, size;
769         size = FFMIN(out_count, s->in_buffer_count);
770         if(size){
771             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
772             ret= swr_convert_internal(s, out, size, &tmp, size);
773             if(ret<0)
774                 return ret;
775             ret2= ret;
776             s->in_buffer_count -= ret;
777             s->in_buffer_index += ret;
778             buf_set(out, out, ret);
779             out_count -= ret;
780             if(!s->in_buffer_count)
781                 s->in_buffer_index = 0;
782         }
783
784         if(in_count){
785             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
786
787             if(in_count > out_count) { //FIXME move after swr_convert_internal
788                 if(   size > s->in_buffer.count
789                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
790                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
791                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
792                     s->in_buffer_index=0;
793                 }else
794                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
795                         return ret;
796             }
797
798             if(out_count){
799                 size = FFMIN(in_count, out_count);
800                 ret= swr_convert_internal(s, out, size, in, size);
801                 if(ret<0)
802                     return ret;
803                 buf_set(in, in, ret);
804                 in_count -= ret;
805                 ret2 += ret;
806             }
807             if(in_count){
808                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
809                 copy(&tmp, in, in_count);
810                 s->in_buffer_count += in_count;
811             }
812         }
813         if(ret2>0 && !s->drop_output)
814             s->outpts += ret2 * (int64_t)s->in_sample_rate;
815         av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
816         return ret2;
817     }
818 }
819
820 int swr_drop_output(struct SwrContext *s, int count){
821     const uint8_t *tmp_arg[SWR_CH_MAX];
822     s->drop_output += count;
823
824     if(s->drop_output <= 0)
825         return 0;
826
827     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
828     return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
829 }
830
831 int swr_inject_silence(struct SwrContext *s, int count){
832     int ret, i;
833     uint8_t *tmp_arg[SWR_CH_MAX];
834
835     if(count <= 0)
836         return 0;
837
838 #define MAX_SILENCE_STEP 16384
839     while (count > MAX_SILENCE_STEP) {
840         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
841             return ret;
842         count -= MAX_SILENCE_STEP;
843     }
844
845     if((ret=swri_realloc_audio(&s->silence, count))<0)
846         return ret;
847
848     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
849         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
850     } else
851         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
852
853     reversefill_audiodata(&s->silence, tmp_arg);
854     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
855     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
856     return ret;
857 }
858
859 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
860     if (s->resampler && s->resample){
861         return s->resampler->get_delay(s, base);
862     }else{
863         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
864     }
865 }
866
867 int swr_get_out_samples(struct SwrContext *s, int in_samples)
868 {
869     int64_t out_samples;
870
871     if (in_samples < 0)
872         return AVERROR(EINVAL);
873
874     if (s->resampler && s->resample) {
875         if (!s->resampler->get_out_samples)
876             return AVERROR(ENOSYS);
877         out_samples = s->resampler->get_out_samples(s, in_samples);
878     } else {
879         out_samples = s->in_buffer_count + in_samples;
880         av_assert0(s->out_sample_rate == s->in_sample_rate);
881     }
882
883     if (out_samples > INT_MAX)
884         return AVERROR(EINVAL);
885
886     return out_samples;
887 }
888
889 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
890     int ret;
891
892     if (!s || compensation_distance < 0)
893         return AVERROR(EINVAL);
894     if (!compensation_distance && sample_delta)
895         return AVERROR(EINVAL);
896     if (!s->resample) {
897         s->flags |= SWR_FLAG_RESAMPLE;
898         ret = swr_init(s);
899         if (ret < 0)
900             return ret;
901     }
902     if (!s->resampler->set_compensation){
903         return AVERROR(EINVAL);
904     }else{
905         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
906     }
907 }
908
909 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
910     if(pts == INT64_MIN)
911         return s->outpts;
912
913     if (s->firstpts == AV_NOPTS_VALUE)
914         s->outpts = s->firstpts = pts;
915
916     if(s->min_compensation >= FLT_MAX) {
917         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
918     } else {
919         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;
920         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
921
922         if(fabs(fdelta) > s->min_compensation) {
923             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
924                 int ret;
925                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
926                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
927                 if(ret<0){
928                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
929                 }
930             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
931                 int duration = s->out_sample_rate * s->soft_compensation_duration;
932                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
933                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
934                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
935                 swr_set_compensation(s, comp, duration);
936             }
937         }
938
939         return s->outpts;
940     }
941 }