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