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