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