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