avcodec/alsdec: Fix integer overflow in decode_var_block_data()
[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     if(s-> in_sample_rate <= 0){
168         av_log(s, AV_LOG_ERROR, "Requested input sample rate %d is invalid\n", s->in_sample_rate);
169         return AVERROR(EINVAL);
170     }
171     if(s->out_sample_rate <= 0){
172         av_log(s, AV_LOG_ERROR, "Requested output sample rate %d is invalid\n", s->out_sample_rate);
173         return AVERROR(EINVAL);
174     }
175     s->out.ch_count  = s-> user_out_ch_count;
176     s-> in.ch_count  = s->  user_in_ch_count;
177     s->used_ch_count = s->user_used_ch_count;
178
179     s-> in_ch_layout = s-> user_in_ch_layout;
180     s->out_ch_layout = s->user_out_ch_layout;
181
182     s->int_sample_fmt= s->user_int_sample_fmt;
183
184     s->dither.method = s->user_dither_method;
185
186     if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
187         av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
188         s->in_ch_layout = 0;
189     }
190
191     if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
192         av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
193         s->out_ch_layout = 0;
194     }
195
196     switch(s->engine){
197 #if CONFIG_LIBSOXR
198         case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
199 #endif
200         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
201         default:
202             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
203             return AVERROR(EINVAL);
204     }
205
206     if(!s->used_ch_count)
207         s->used_ch_count= s->in.ch_count;
208
209     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
210         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
211         s-> in_ch_layout= 0;
212     }
213
214     if(!s-> in_ch_layout)
215         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
216     if(!s->out_ch_layout)
217         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
218
219     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
220                  s->rematrix_custom;
221
222     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
223         if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
224            && av_get_bytes_per_sample(s->out_sample_fmt) <= 2){
225             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
226         }else if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
227            && !s->rematrix
228            && s->out_sample_rate==s->in_sample_rate
229            && !(s->flags & SWR_FLAG_RESAMPLE)){
230             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
231         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
232                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
233                  && !s->rematrix
234                  && s->out_sample_rate == s->in_sample_rate
235                  && !(s->flags & SWR_FLAG_RESAMPLE)
236                  && s->engine != SWR_ENGINE_SOXR){
237             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
238         }else if(av_get_bytes_per_sample(s->in_sample_fmt) <= 4){
239             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
240         }else{
241             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
242         }
243     }
244     av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
245
246     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
247         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
248         &&s->int_sample_fmt != AV_SAMPLE_FMT_S64P
249         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
250         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
251         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, s16p/s32p/s64p/fltp/dblp are supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
252         return AVERROR(EINVAL);
253     }
254
255     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
256     set_audiodata_fmt(&s->out, s->out_sample_fmt);
257
258     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
259         if (!s->async && s->min_compensation >= FLT_MAX/2)
260             s->async = 1;
261         s->firstpts =
262         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
263     } else
264         s->firstpts = AV_NOPTS_VALUE;
265
266     if (s->async) {
267         if (s->min_compensation >= FLT_MAX/2)
268             s->min_compensation = 0.001;
269         if (s->async > 1.0001) {
270             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
271         }
272     }
273
274     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
275         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);
276         if (!s->resample) {
277             av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
278             return AVERROR(ENOMEM);
279         }
280     }else
281         s->resampler->free(&s->resample);
282     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
283         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
284         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
285         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
286         && s->resample){
287         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16p/s32p/fltp/dblp\n");
288         ret = AVERROR(EINVAL);
289         goto fail;
290     }
291
292 #define RSC 1 //FIXME finetune
293     if(!s-> in.ch_count)
294         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
295     if(!s->used_ch_count)
296         s->used_ch_count= s->in.ch_count;
297     if(!s->out.ch_count)
298         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
299
300     if(!s-> in.ch_count){
301         av_assert0(!s->in_ch_layout);
302         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
303         ret = AVERROR(EINVAL);
304         goto fail;
305     }
306
307     av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
308     av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
309     if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
310         av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
311         ret = AVERROR(EINVAL);
312         goto fail;
313     }
314     if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
315         av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
316         ret = AVERROR(EINVAL);
317         goto fail;
318     }
319
320     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
321         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
322                "but there is not enough information to do it\n", l1, l2);
323         ret = AVERROR(EINVAL);
324         goto fail;
325     }
326
327 av_assert0(s->used_ch_count);
328 av_assert0(s->out.ch_count);
329     s->resample_first= RSC*s->out.ch_count/s->used_ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
330
331     s->in_buffer= s->in;
332     s->silence  = s->in;
333     s->drop_temp= s->out;
334
335     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
336         goto fail;
337
338     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
339         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
340                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
341         return 0;
342     }
343
344     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
345                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
346     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
347                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
348
349     if (!s->in_convert || !s->out_convert) {
350         ret = AVERROR(ENOMEM);
351         goto fail;
352     }
353
354     s->postin= s->in;
355     s->preout= s->out;
356     s->midbuf= s->in;
357
358     if(s->channel_map){
359         s->postin.ch_count=
360         s->midbuf.ch_count= s->used_ch_count;
361         if(s->resample)
362             s->in_buffer.ch_count= s->used_ch_count;
363     }
364     if(!s->resample_first){
365         s->midbuf.ch_count= s->out.ch_count;
366         if(s->resample)
367             s->in_buffer.ch_count = s->out.ch_count;
368     }
369
370     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
371     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
372     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
373
374     if(s->resample){
375         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
376     }
377
378     av_assert0(!s->preout.count);
379     s->dither.noise = s->preout;
380     s->dither.temp  = s->preout;
381     if (s->dither.method > SWR_DITHER_NS) {
382         s->dither.noise.bps = 4;
383         s->dither.noise.fmt = AV_SAMPLE_FMT_FLTP;
384         s->dither.noise_scale = 1;
385     }
386
387     if(s->rematrix || s->dither.method) {
388         ret = swri_rematrix_init(s);
389         if (ret < 0)
390             goto fail;
391     }
392
393     return 0;
394 fail:
395     swr_close(s);
396     return ret;
397
398 }
399
400 int swri_realloc_audio(AudioData *a, int count){
401     int i, countb;
402     AudioData old;
403
404     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
405         return AVERROR(EINVAL);
406
407     if(a->count >= count)
408         return 0;
409
410     count*=2;
411
412     countb= FFALIGN(count*a->bps, ALIGN);
413     old= *a;
414
415     av_assert0(a->bps);
416     av_assert0(a->ch_count);
417
418     a->data= av_mallocz_array(countb, a->ch_count);
419     if(!a->data)
420         return AVERROR(ENOMEM);
421     for(i=0; i<a->ch_count; i++){
422         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
423         if(a->count && a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
424     }
425     if(a->count && !a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
426     av_freep(&old.data);
427     a->count= count;
428
429     return 1;
430 }
431
432 static void copy(AudioData *out, AudioData *in,
433                  int count){
434     av_assert0(out->planar == in->planar);
435     av_assert0(out->bps == in->bps);
436     av_assert0(out->ch_count == in->ch_count);
437     if(out->planar){
438         int ch;
439         for(ch=0; ch<out->ch_count; ch++)
440             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
441     }else
442         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
443 }
444
445 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
446     int i;
447     if(!in_arg){
448         memset(out->ch, 0, sizeof(out->ch));
449     }else if(out->planar){
450         for(i=0; i<out->ch_count; i++)
451             out->ch[i]= in_arg[i];
452     }else{
453         for(i=0; i<out->ch_count; i++)
454             out->ch[i]= in_arg[0] + i*out->bps;
455     }
456 }
457
458 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
459     int i;
460     if(out->planar){
461         for(i=0; i<out->ch_count; i++)
462             in_arg[i]= out->ch[i];
463     }else{
464         in_arg[0]= out->ch[0];
465     }
466 }
467
468 /**
469  *
470  * out may be equal in.
471  */
472 static void buf_set(AudioData *out, AudioData *in, int count){
473     int ch;
474     if(in->planar){
475         for(ch=0; ch<out->ch_count; ch++)
476             out->ch[ch]= in->ch[ch] + count*out->bps;
477     }else{
478         for(ch=out->ch_count-1; ch>=0; ch--)
479             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
480     }
481 }
482
483 /**
484  *
485  * @return number of samples output per channel
486  */
487 static int resample(SwrContext *s, AudioData *out_param, int out_count,
488                              const AudioData * in_param, int in_count){
489     AudioData in, out, tmp;
490     int ret_sum=0;
491     int border=0;
492     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
493
494     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
495     av_assert1(s->in_buffer.planar   == in_param->planar);
496     av_assert1(s->in_buffer.fmt      == in_param->fmt);
497
498     tmp=out=*out_param;
499     in =  *in_param;
500
501     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
502                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
503     if (border == INT_MAX) {
504         return 0;
505     } else if (border < 0) {
506         return border;
507     } else if (border) {
508         buf_set(&in, &in, border);
509         in_count -= border;
510         s->resample_in_constraint = 0;
511     }
512
513     do{
514         int ret, size, consumed;
515         if(!s->resample_in_constraint && s->in_buffer_count){
516             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
517             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
518             out_count -= ret;
519             ret_sum += ret;
520             buf_set(&out, &out, ret);
521             s->in_buffer_count -= consumed;
522             s->in_buffer_index += consumed;
523
524             if(!in_count)
525                 break;
526             if(s->in_buffer_count <= border){
527                 buf_set(&in, &in, -s->in_buffer_count);
528                 in_count += s->in_buffer_count;
529                 s->in_buffer_count=0;
530                 s->in_buffer_index=0;
531                 border = 0;
532             }
533         }
534
535         if((s->flushed || in_count > padless) && !s->in_buffer_count){
536             s->in_buffer_index=0;
537             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
538             out_count -= ret;
539             ret_sum += ret;
540             buf_set(&out, &out, ret);
541             in_count -= consumed;
542             buf_set(&in, &in, consumed);
543         }
544
545         //TODO is this check sane considering the advanced copy avoidance below
546         size= s->in_buffer_index + s->in_buffer_count + in_count;
547         if(   size > s->in_buffer.count
548            && s->in_buffer_count + in_count <= s->in_buffer_index){
549             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
550             copy(&s->in_buffer, &tmp, s->in_buffer_count);
551             s->in_buffer_index=0;
552         }else
553             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
554                 return ret;
555
556         if(in_count){
557             int count= in_count;
558             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
559
560             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
561             copy(&tmp, &in, /*in_*/count);
562             s->in_buffer_count += count;
563             in_count -= count;
564             border += count;
565             buf_set(&in, &in, count);
566             s->resample_in_constraint= 0;
567             if(s->in_buffer_count != count || in_count)
568                 continue;
569             if (padless) {
570                 padless = 0;
571                 continue;
572             }
573         }
574         break;
575     }while(1);
576
577     s->resample_in_constraint= !!out_count;
578
579     return ret_sum;
580 }
581
582 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
583                                                       AudioData *in , int  in_count){
584     AudioData *postin, *midbuf, *preout;
585     int ret/*, in_max*/;
586     AudioData preout_tmp, midbuf_tmp;
587
588     if(s->full_convert){
589         av_assert0(!s->resample);
590         swri_audio_convert(s->full_convert, out, in, in_count);
591         return out_count;
592     }
593
594 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
595 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
596
597     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
598         return ret;
599     if(s->resample_first){
600         av_assert0(s->midbuf.ch_count == s->used_ch_count);
601         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
602             return ret;
603     }else{
604         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
605         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
606             return ret;
607     }
608     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
609         return ret;
610
611     postin= &s->postin;
612
613     midbuf_tmp= s->midbuf;
614     midbuf= &midbuf_tmp;
615     preout_tmp= s->preout;
616     preout= &preout_tmp;
617
618     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
619         postin= in;
620
621     if(s->resample_first ? !s->resample : !s->rematrix)
622         midbuf= postin;
623
624     if(s->resample_first ? !s->rematrix : !s->resample)
625         preout= midbuf;
626
627     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
628        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
629         if(preout==in){
630             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
631             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
632             copy(out, in, out_count);
633             return out_count;
634         }
635         else if(preout==postin) preout= midbuf= postin= out;
636         else if(preout==midbuf) preout= midbuf= out;
637         else                    preout= out;
638     }
639
640     if(in != postin){
641         swri_audio_convert(s->in_convert, postin, in, in_count);
642     }
643
644     if(s->resample_first){
645         if(postin != midbuf)
646             out_count= resample(s, midbuf, out_count, postin, in_count);
647         if(midbuf != preout)
648             swri_rematrix(s, preout, midbuf, out_count, preout==out);
649     }else{
650         if(postin != midbuf)
651             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
652         if(midbuf != preout)
653             out_count= resample(s, preout, out_count, midbuf, in_count);
654     }
655
656     if(preout != out && out_count){
657         AudioData *conv_src = preout;
658         if(s->dither.method){
659             int ch;
660             int dither_count= FFMAX(out_count, 1<<16);
661
662             if (preout == in) {
663                 conv_src = &s->dither.temp;
664                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
665                     return ret;
666             }
667
668             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
669                 return ret;
670             if(ret)
671                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
672                     if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
673                         return ret;
674             av_assert0(s->dither.noise.ch_count == preout->ch_count);
675
676             if(s->dither.noise_pos + out_count > s->dither.noise.count)
677                 s->dither.noise_pos = 0;
678
679             if (s->dither.method < SWR_DITHER_NS){
680                 if (s->mix_2_1_simd) {
681                     int len1= out_count&~15;
682                     int off = len1 * preout->bps;
683
684                     if(len1)
685                         for(ch=0; ch<preout->ch_count; ch++)
686                             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);
687                     if(out_count != len1)
688                         for(ch=0; ch<preout->ch_count; ch++)
689                             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, s->native_one, 0, 0, out_count - len1);
690                 } else {
691                     for(ch=0; ch<preout->ch_count; ch++)
692                         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);
693                 }
694             } else {
695                 switch(s->int_sample_fmt) {
696                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
697                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
698                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
699                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
700                 }
701             }
702             s->dither.noise_pos += out_count;
703         }
704 //FIXME packed doesn't need more than 1 chan here!
705         swri_audio_convert(s->out_convert, out, conv_src, out_count);
706     }
707     return out_count;
708 }
709
710 int swr_is_initialized(struct SwrContext *s) {
711     return !!s->in_buffer.ch_count;
712 }
713
714 int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
715                                                     const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
716     AudioData * in= &s->in;
717     AudioData *out= &s->out;
718     int av_unused max_output;
719
720     if (!swr_is_initialized(s)) {
721         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
722         return AVERROR(EINVAL);
723     }
724 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
725     max_output = swr_get_out_samples(s, in_count);
726 #endif
727
728     while(s->drop_output > 0){
729         int ret;
730         uint8_t *tmp_arg[SWR_CH_MAX];
731 #define MAX_DROP_STEP 16384
732         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
733             return ret;
734
735         reversefill_audiodata(&s->drop_temp, tmp_arg);
736         s->drop_output *= -1; //FIXME find a less hackish solution
737         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
738         s->drop_output *= -1;
739         in_count = 0;
740         if(ret>0) {
741             s->drop_output -= ret;
742             if (!s->drop_output && !out_arg)
743                 return 0;
744             continue;
745         }
746
747         av_assert0(s->drop_output);
748         return 0;
749     }
750
751     if(!in_arg){
752         if(s->resample){
753             if (!s->flushed)
754                 s->resampler->flush(s);
755             s->resample_in_constraint = 0;
756             s->flushed = 1;
757         }else if(!s->in_buffer_count){
758             return 0;
759         }
760     }else
761         fill_audiodata(in ,  (void*)in_arg);
762
763     fill_audiodata(out, out_arg);
764
765     if(s->resample){
766         int ret = swr_convert_internal(s, out, out_count, in, in_count);
767         if(ret>0 && !s->drop_output)
768             s->outpts += ret * (int64_t)s->in_sample_rate;
769
770         av_assert2(max_output < 0 || ret < 0 || ret <= max_output);
771
772         return ret;
773     }else{
774         AudioData tmp= *in;
775         int ret2=0;
776         int ret, size;
777         size = FFMIN(out_count, s->in_buffer_count);
778         if(size){
779             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
780             ret= swr_convert_internal(s, out, size, &tmp, size);
781             if(ret<0)
782                 return ret;
783             ret2= ret;
784             s->in_buffer_count -= ret;
785             s->in_buffer_index += ret;
786             buf_set(out, out, ret);
787             out_count -= ret;
788             if(!s->in_buffer_count)
789                 s->in_buffer_index = 0;
790         }
791
792         if(in_count){
793             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
794
795             if(in_count > out_count) { //FIXME move after swr_convert_internal
796                 if(   size > s->in_buffer.count
797                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
798                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
799                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
800                     s->in_buffer_index=0;
801                 }else
802                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
803                         return ret;
804             }
805
806             if(out_count){
807                 size = FFMIN(in_count, out_count);
808                 ret= swr_convert_internal(s, out, size, in, size);
809                 if(ret<0)
810                     return ret;
811                 buf_set(in, in, ret);
812                 in_count -= ret;
813                 ret2 += ret;
814             }
815             if(in_count){
816                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
817                 copy(&tmp, in, in_count);
818                 s->in_buffer_count += in_count;
819             }
820         }
821         if(ret2>0 && !s->drop_output)
822             s->outpts += ret2 * (int64_t)s->in_sample_rate;
823         av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
824         return ret2;
825     }
826 }
827
828 int swr_drop_output(struct SwrContext *s, int count){
829     const uint8_t *tmp_arg[SWR_CH_MAX];
830     s->drop_output += count;
831
832     if(s->drop_output <= 0)
833         return 0;
834
835     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
836     return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
837 }
838
839 int swr_inject_silence(struct SwrContext *s, int count){
840     int ret, i;
841     uint8_t *tmp_arg[SWR_CH_MAX];
842
843     if(count <= 0)
844         return 0;
845
846 #define MAX_SILENCE_STEP 16384
847     while (count > MAX_SILENCE_STEP) {
848         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
849             return ret;
850         count -= MAX_SILENCE_STEP;
851     }
852
853     if((ret=swri_realloc_audio(&s->silence, count))<0)
854         return ret;
855
856     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
857         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
858     } else
859         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
860
861     reversefill_audiodata(&s->silence, tmp_arg);
862     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
863     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
864     return ret;
865 }
866
867 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
868     if (s->resampler && s->resample){
869         return s->resampler->get_delay(s, base);
870     }else{
871         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
872     }
873 }
874
875 int swr_get_out_samples(struct SwrContext *s, int in_samples)
876 {
877     int64_t out_samples;
878
879     if (in_samples < 0)
880         return AVERROR(EINVAL);
881
882     if (s->resampler && s->resample) {
883         if (!s->resampler->get_out_samples)
884             return AVERROR(ENOSYS);
885         out_samples = s->resampler->get_out_samples(s, in_samples);
886     } else {
887         out_samples = s->in_buffer_count + in_samples;
888         av_assert0(s->out_sample_rate == s->in_sample_rate);
889     }
890
891     if (out_samples > INT_MAX)
892         return AVERROR(EINVAL);
893
894     return out_samples;
895 }
896
897 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
898     int ret;
899
900     if (!s || compensation_distance < 0)
901         return AVERROR(EINVAL);
902     if (!compensation_distance && sample_delta)
903         return AVERROR(EINVAL);
904     if (!s->resample) {
905         s->flags |= SWR_FLAG_RESAMPLE;
906         ret = swr_init(s);
907         if (ret < 0)
908             return ret;
909     }
910     if (!s->resampler->set_compensation){
911         return AVERROR(EINVAL);
912     }else{
913         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
914     }
915 }
916
917 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
918     if(pts == INT64_MIN)
919         return s->outpts;
920
921     if (s->firstpts == AV_NOPTS_VALUE)
922         s->outpts = s->firstpts = pts;
923
924     if(s->min_compensation >= FLT_MAX) {
925         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
926     } else {
927         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;
928         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
929
930         if(fabs(fdelta) > s->min_compensation) {
931             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
932                 int ret;
933                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
934                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
935                 if(ret<0){
936                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
937                 }
938             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
939                 int duration = s->out_sample_rate * s->soft_compensation_duration;
940                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
941                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
942                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
943                 swr_set_compensation(s, comp, duration);
944             }
945         }
946
947         return s->outpts;
948     }
949 }