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 #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, "tsf", AV_SAMPLE_FMT_NONE,   0) < 0)
  88         goto fail;
  89
  90     if (av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> user_in_ch_layout), 0) < 0)
  91         goto fail;
  92
  93     if (av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->user_out_ch_layout), 0) < 0)
  94         goto fail;
  95
  96     av_opt_set_int(s, "uch", 0, 0);
  97     return s;
  98 fail:
  99     av_log(s, AV_LOG_ERROR, "Failed to set option\n");
 100     swr_free(&s);
 101     return NULL;
 102 }
 103
 104 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 105     a->fmt   = fmt;
 106     a->bps   = av_get_bytes_per_sample(fmt);
 107     a->planar= av_sample_fmt_is_planar(fmt);
 108     if (a->ch_count == 1)
 109         a->planar = 1;
 110 }
 111
 112 static void free_temp(AudioData *a){
 113     av_free(a->data);
 114     memset(a, 0, sizeof(*a));
 115 }
 116
 117 static void clear_context(SwrContext *s){
 118     s->in_buffer_index= 0;
 119     s->in_buffer_count= 0;
 120     s->resample_in_constraint= 0;
 121     memset(s->in.ch, 0, sizeof(s->in.ch));
 122     memset(s->out.ch, 0, sizeof(s->out.ch));
 123     free_temp(&s->postin);
 124     free_temp(&s->midbuf);
 125     free_temp(&s->preout);
 126     free_temp(&s->in_buffer);
 127     free_temp(&s->silence);
 128     free_temp(&s->drop_temp);
 129     free_temp(&s->dither.noise);
 130     free_temp(&s->dither.temp);
 131     swri_audio_convert_free(&s-> in_convert);
 132     swri_audio_convert_free(&s->out_convert);
 133     swri_audio_convert_free(&s->full_convert);
 134     swri_rematrix_free(s);
 135
 136     s->flushed = 0;
 137 }
 138
 139 av_cold void swr_free(SwrContext **ss){
 140     SwrContext *s= *ss;
 141     if(s){
 142         clear_context(s);
 143         if (s->resampler)
 144             s->resampler->free(&s->resample);
 145     }
 146
 147     av_freep(ss);
 148 }
 149
 150 av_cold void swr_close(SwrContext *s){
 151     clear_context(s);
 152 }
 153
 154 av_cold int swr_init(struct SwrContext *s){
 155     int ret;
 156     char l1[1024], l2[1024];
 157
 158     clear_context(s);
 159
 160     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 161         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 162         return AVERROR(EINVAL);
 163     }
 164     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 165         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 166         return AVERROR(EINVAL);
 167     }
 168
 169     s->out.ch_count  = s-> user_out_ch_count;
 170     s-> in.ch_count  = s->  user_in_ch_count;
 171     s->used_ch_count = s->user_used_ch_count;
 172
 173     s-> in_ch_layout = s-> user_in_ch_layout;
 174     s->out_ch_layout = s->user_out_ch_layout;
 175
 176     if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
 177         av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
 178         s->in_ch_layout = 0;
 179     }
 180
 181     if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
 182         av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
 183         s->out_ch_layout = 0;
 184     }
 185
 186     switch(s->engine){
 187 #if CONFIG_LIBSOXR
 188         case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
 189 #endif
 190         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
 191         default:
 192             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
 193             return AVERROR(EINVAL);
 194     }
 195
 196     if(!s->used_ch_count)
 197         s->used_ch_count= s->in.ch_count;
 198
 199     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 200         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 201         s-> in_ch_layout= 0;
 202     }
 203
 204     if(!s-> in_ch_layout)
 205         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 206     if(!s->out_ch_layout)
 207         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 208
 209     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 210                  s->rematrix_custom;
 211
 212     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 213         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 214             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 215         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
 216                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
 217                  && !s->rematrix
 218                  && s->engine != SWR_ENGINE_SOXR){
 219             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
 220         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 221             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 222         }else{
 223             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 224             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 225         }
 226     }
 227
 228     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 229         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 230         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 231         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 232         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));
 233         return AVERROR(EINVAL);
 234     }
 235
 236     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 237     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 238
 239     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
 240         if (!s->async && s->min_compensation >= FLT_MAX/2)
 241             s->async = 1;
 242         s->firstpts =
 243         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
 244     } else
 245         s->firstpts = AV_NOPTS_VALUE;
 246
 247     if (s->async) {
 248         if (s->min_compensation >= FLT_MAX/2)
 249             s->min_compensation = 0.001;
 250         if (s->async > 1.0001) {
 251             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
 252         }
 253     }
 254
 255     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 256         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);
 257         if (!s->resample) {
 258             av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
 259             return AVERROR(ENOMEM);
 260         }
 261     }else
 262         s->resampler->free(&s->resample);
 263     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 264         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 265         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 266         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 267         && s->resample){
 268         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 269         ret = AVERROR(EINVAL);
 270         goto fail;
 271     }
 272
 273 #define RSC 1 //FIXME finetune
 274     if(!s-> in.ch_count)
 275         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 276     if(!s->used_ch_count)
 277         s->used_ch_count= s->in.ch_count;
 278     if(!s->out.ch_count)
 279         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 280
 281     if(!s-> in.ch_count){
 282         av_assert0(!s->in_ch_layout);
 283         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 284         ret = AVERROR(EINVAL);
 285         goto fail;
 286     }
 287
 288     av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
 289     av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
 290     if (s->out_ch_layout && s->out.ch_count != av_get_channel_layout_nb_channels(s->out_ch_layout)) {
 291         av_log(s, AV_LOG_ERROR, "Output channel layout %s mismatches specified channel count %d\n", l2, s->out.ch_count);
 292         ret = AVERROR(EINVAL);
 293         goto fail;
 294     }
 295     if (s->in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s->in_ch_layout)) {
 296         av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_count);
 297         ret = AVERROR(EINVAL);
 298         goto fail;
 299     }
 300
 301     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 302         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
 303                "but there is not enough information to do it\n", l1, l2);
 304         ret = AVERROR(EINVAL);
 305         goto fail;
 306     }
 307
 308 av_assert0(s->used_ch_count);
 309 av_assert0(s->out.ch_count);
 310     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 311
 312     s->in_buffer= s->in;
 313     s->silence  = s->in;
 314     s->drop_temp= s->out;
 315
 316     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
 317         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 318                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 319         return 0;
 320     }
 321
 322     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 323                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 324     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 325                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 326
 327     if (!s->in_convert || !s->out_convert) {
 328         ret = AVERROR(ENOMEM);
 329         goto fail;
 330     }
 331
 332     s->postin= s->in;
 333     s->preout= s->out;
 334     s->midbuf= s->in;
 335
 336     if(s->channel_map){
 337         s->postin.ch_count=
 338         s->midbuf.ch_count= s->used_ch_count;
 339         if(s->resample)
 340             s->in_buffer.ch_count= s->used_ch_count;
 341     }
 342     if(!s->resample_first){
 343         s->midbuf.ch_count= s->out.ch_count;
 344         if(s->resample)
 345             s->in_buffer.ch_count = s->out.ch_count;
 346     }
 347
 348     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 349     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 350     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 351
 352     if(s->resample){
 353         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 354     }
 355
 356     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
 357         goto fail;
 358
 359     if(s->rematrix || s->dither.method) {
 360         ret = swri_rematrix_init(s);
 361         if (ret < 0)
 362             goto fail;
 363     }
 364
 365     return 0;
 366 fail:
 367     swr_close(s);
 368     return ret;
 369
 370 }
 371
 372 int swri_realloc_audio(AudioData *a, int count){
 373     int i, countb;
 374     AudioData old;
 375
 376     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 377         return AVERROR(EINVAL);
 378
 379     if(a->count >= count)
 380         return 0;
 381
 382     count*=2;
 383
 384     countb= FFALIGN(count*a->bps, ALIGN);
 385     old= *a;
 386
 387     av_assert0(a->bps);
 388     av_assert0(a->ch_count);
 389
 390     a->data= av_mallocz_array(countb, a->ch_count);
 391     if(!a->data)
 392         return AVERROR(ENOMEM);
 393     for(i=0; i<a->ch_count; i++){
 394         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 395         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 396     }
 397     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 398     av_freep(&old.data);
 399     a->count= count;
 400
 401     return 1;
 402 }
 403
 404 static void copy(AudioData *out, AudioData *in,
 405                  int count){
 406     av_assert0(out->planar == in->planar);
 407     av_assert0(out->bps == in->bps);
 408     av_assert0(out->ch_count == in->ch_count);
 409     if(out->planar){
 410         int ch;
 411         for(ch=0; ch<out->ch_count; ch++)
 412             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 413     }else
 414         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 415 }
 416
 417 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 418     int i;
 419     if(!in_arg){
 420         memset(out->ch, 0, sizeof(out->ch));
 421     }else if(out->planar){
 422         for(i=0; i<out->ch_count; i++)
 423             out->ch[i]= in_arg[i];
 424     }else{
 425         for(i=0; i<out->ch_count; i++)
 426             out->ch[i]= in_arg[0] + i*out->bps;
 427     }
 428 }
 429
 430 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 431     int i;
 432     if(out->planar){
 433         for(i=0; i<out->ch_count; i++)
 434             in_arg[i]= out->ch[i];
 435     }else{
 436         in_arg[0]= out->ch[0];
 437     }
 438 }
 439
 440 /**
 441  *
 442  * out may be equal in.
 443  */
 444 static void buf_set(AudioData *out, AudioData *in, int count){
 445     int ch;
 446     if(in->planar){
 447         for(ch=0; ch<out->ch_count; ch++)
 448             out->ch[ch]= in->ch[ch] + count*out->bps;
 449     }else{
 450         for(ch=out->ch_count-1; ch>=0; ch--)
 451             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 452     }
 453 }
 454
 455 /**
 456  *
 457  * @return number of samples output per channel
 458  */
 459 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 460                              const AudioData * in_param, int in_count){
 461     AudioData in, out, tmp;
 462     int ret_sum=0;
 463     int border=0;
 464     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
 465
 466     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 467     av_assert1(s->in_buffer.planar   == in_param->planar);
 468     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 469
 470     tmp=out=*out_param;
 471     in =  *in_param;
 472
 473     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
 474                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
 475     if (border == INT_MAX) {
 476         return 0;
 477     } else if (border < 0) {
 478         return border;
 479     } else if (border) {
 480         buf_set(&in, &in, border);
 481         in_count -= border;
 482         s->resample_in_constraint = 0;
 483     }
 484
 485     do{
 486         int ret, size, consumed;
 487         if(!s->resample_in_constraint && s->in_buffer_count){
 488             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 489             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 490             out_count -= ret;
 491             ret_sum += ret;
 492             buf_set(&out, &out, ret);
 493             s->in_buffer_count -= consumed;
 494             s->in_buffer_index += consumed;
 495
 496             if(!in_count)
 497                 break;
 498             if(s->in_buffer_count <= border){
 499                 buf_set(&in, &in, -s->in_buffer_count);
 500                 in_count += s->in_buffer_count;
 501                 s->in_buffer_count=0;
 502                 s->in_buffer_index=0;
 503                 border = 0;
 504             }
 505         }
 506
 507         if((s->flushed || in_count > padless) && !s->in_buffer_count){
 508             s->in_buffer_index=0;
 509             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
 510             out_count -= ret;
 511             ret_sum += ret;
 512             buf_set(&out, &out, ret);
 513             in_count -= consumed;
 514             buf_set(&in, &in, consumed);
 515         }
 516
 517         //TODO is this check sane considering the advanced copy avoidance below
 518         size= s->in_buffer_index + s->in_buffer_count + in_count;
 519         if(   size > s->in_buffer.count
 520            && s->in_buffer_count + in_count <= s->in_buffer_index){
 521             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 522             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 523             s->in_buffer_index=0;
 524         }else
 525             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 526                 return ret;
 527
 528         if(in_count){
 529             int count= in_count;
 530             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 531
 532             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 533             copy(&tmp, &in, /*in_*/count);
 534             s->in_buffer_count += count;
 535             in_count -= count;
 536             border += count;
 537             buf_set(&in, &in, count);
 538             s->resample_in_constraint= 0;
 539             if(s->in_buffer_count != count || in_count)
 540                 continue;
 541             if (padless) {
 542                 padless = 0;
 543                 continue;
 544             }
 545         }
 546         break;
 547     }while(1);
 548
 549     s->resample_in_constraint= !!out_count;
 550
 551     return ret_sum;
 552 }
 553
 554 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 555                                                       AudioData *in , int  in_count){
 556     AudioData *postin, *midbuf, *preout;
 557     int ret/*, in_max*/;
 558     AudioData preout_tmp, midbuf_tmp;
 559
 560     if(s->full_convert){
 561         av_assert0(!s->resample);
 562         swri_audio_convert(s->full_convert, out, in, in_count);
 563         return out_count;
 564     }
 565
 566 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 567 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 568
 569     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
 570         return ret;
 571     if(s->resample_first){
 572         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 573         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
 574             return ret;
 575     }else{
 576         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 577         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
 578             return ret;
 579     }
 580     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
 581         return ret;
 582
 583     postin= &s->postin;
 584
 585     midbuf_tmp= s->midbuf;
 586     midbuf= &midbuf_tmp;
 587     preout_tmp= s->preout;
 588     preout= &preout_tmp;
 589
 590     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 591         postin= in;
 592
 593     if(s->resample_first ? !s->resample : !s->rematrix)
 594         midbuf= postin;
 595
 596     if(s->resample_first ? !s->rematrix : !s->resample)
 597         preout= midbuf;
 598
 599     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
 600        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
 601         if(preout==in){
 602             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 603             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 604             copy(out, in, out_count);
 605             return out_count;
 606         }
 607         else if(preout==postin) preout= midbuf= postin= out;
 608         else if(preout==midbuf) preout= midbuf= out;
 609         else                    preout= out;
 610     }
 611
 612     if(in != postin){
 613         swri_audio_convert(s->in_convert, postin, in, in_count);
 614     }
 615
 616     if(s->resample_first){
 617         if(postin != midbuf)
 618             out_count= resample(s, midbuf, out_count, postin, in_count);
 619         if(midbuf != preout)
 620             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 621     }else{
 622         if(postin != midbuf)
 623             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 624         if(midbuf != preout)
 625             out_count= resample(s, preout, out_count, midbuf, in_count);
 626     }
 627
 628     if(preout != out && out_count){
 629         AudioData *conv_src = preout;
 630         if(s->dither.method){
 631             int ch;
 632             int dither_count= FFMAX(out_count, 1<<16);
 633
 634             if (preout == in) {
 635                 conv_src = &s->dither.temp;
 636                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
 637                     return ret;
 638             }
 639
 640             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
 641                 return ret;
 642             if(ret)
 643                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
 644                     if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt))<0)
 645                         return ret;
 646             av_assert0(s->dither.noise.ch_count == preout->ch_count);
 647
 648             if(s->dither.noise_pos + out_count > s->dither.noise.count)
 649                 s->dither.noise_pos = 0;
 650
 651             if (s->dither.method < SWR_DITHER_NS){
 652                 if (s->mix_2_1_simd) {
 653                     int len1= out_count&~15;
 654                     int off = len1 * preout->bps;
 655
 656                     if(len1)
 657                         for(ch=0; ch<preout->ch_count; ch++)
 658                             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);
 659                     if(out_count != len1)
 660                         for(ch=0; ch<preout->ch_count; ch++)
 661                             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);
 662                 } else {
 663                     for(ch=0; ch<preout->ch_count; ch++)
 664                         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);
 665                 }
 666             } else {
 667                 switch(s->int_sample_fmt) {
 668                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
 669                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
 670                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
 671                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
 672                 }
 673             }
 674             s->dither.noise_pos += out_count;
 675         }
 676 //FIXME packed doesn't need more than 1 chan here!
 677         swri_audio_convert(s->out_convert, out, conv_src, out_count);
 678     }
 679     return out_count;
 680 }
 681
 682 int swr_is_initialized(struct SwrContext *s) {
 683     return !!s->in_buffer.ch_count;
 684 }
 685
 686 int attribute_align_arg swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 687                                                     const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 688     AudioData * in= &s->in;
 689     AudioData *out= &s->out;
 690     int av_unused max_output;
 691
 692     if (!swr_is_initialized(s)) {
 693         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
 694         return AVERROR(EINVAL);
 695     }
 696 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
 697     max_output = swr_get_out_samples(s, in_count);
 698 #endif
 699
 700     while(s->drop_output > 0){
 701         int ret;
 702         uint8_t *tmp_arg[SWR_CH_MAX];
 703 #define MAX_DROP_STEP 16384
 704         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
 705             return ret;
 706
 707         reversefill_audiodata(&s->drop_temp, tmp_arg);
 708         s->drop_output *= -1; //FIXME find a less hackish solution
 709         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
 710         s->drop_output *= -1;
 711         in_count = 0;
 712         if(ret>0) {
 713             s->drop_output -= ret;
 714             if (!s->drop_output && !out_arg)
 715                 return 0;
 716             continue;
 717         }
 718
 719         av_assert0(s->drop_output);
 720         return 0;
 721     }
 722
 723     if(!in_arg){
 724         if(s->resample){
 725             if (!s->flushed)
 726                 s->resampler->flush(s);
 727             s->resample_in_constraint = 0;
 728             s->flushed = 1;
 729         }else if(!s->in_buffer_count){
 730             return 0;
 731         }
 732     }else
 733         fill_audiodata(in ,  (void*)in_arg);
 734
 735     fill_audiodata(out, out_arg);
 736
 737     if(s->resample){
 738         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 739         if(ret>0 && !s->drop_output)
 740             s->outpts += ret * (int64_t)s->in_sample_rate;
 741
 742         av_assert2(max_output < 0 || ret < 0 || ret <= max_output);
 743
 744         return ret;
 745     }else{
 746         AudioData tmp= *in;
 747         int ret2=0;
 748         int ret, size;
 749         size = FFMIN(out_count, s->in_buffer_count);
 750         if(size){
 751             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 752             ret= swr_convert_internal(s, out, size, &tmp, size);
 753             if(ret<0)
 754                 return ret;
 755             ret2= ret;
 756             s->in_buffer_count -= ret;
 757             s->in_buffer_index += ret;
 758             buf_set(out, out, ret);
 759             out_count -= ret;
 760             if(!s->in_buffer_count)
 761                 s->in_buffer_index = 0;
 762         }
 763
 764         if(in_count){
 765             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 766
 767             if(in_count > out_count) { //FIXME move after swr_convert_internal
 768                 if(   size > s->in_buffer.count
 769                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 770                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 771                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 772                     s->in_buffer_index=0;
 773                 }else
 774                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 775                         return ret;
 776             }
 777
 778             if(out_count){
 779                 size = FFMIN(in_count, out_count);
 780                 ret= swr_convert_internal(s, out, size, in, size);
 781                 if(ret<0)
 782                     return ret;
 783                 buf_set(in, in, ret);
 784                 in_count -= ret;
 785                 ret2 += ret;
 786             }
 787             if(in_count){
 788                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 789                 copy(&tmp, in, in_count);
 790                 s->in_buffer_count += in_count;
 791             }
 792         }
 793         if(ret2>0 && !s->drop_output)
 794             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 795         av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
 796         return ret2;
 797     }
 798 }
 799
 800 int swr_drop_output(struct SwrContext *s, int count){
 801     const uint8_t *tmp_arg[SWR_CH_MAX];
 802     s->drop_output += count;
 803
 804     if(s->drop_output <= 0)
 805         return 0;
 806
 807     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 808     return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
 809 }
 810
 811 int swr_inject_silence(struct SwrContext *s, int count){
 812     int ret, i;
 813     uint8_t *tmp_arg[SWR_CH_MAX];
 814
 815     if(count <= 0)
 816         return 0;
 817
 818 #define MAX_SILENCE_STEP 16384
 819     while (count > MAX_SILENCE_STEP) {
 820         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
 821             return ret;
 822         count -= MAX_SILENCE_STEP;
 823     }
 824
 825     if((ret=swri_realloc_audio(&s->silence, count))<0)
 826         return ret;
 827
 828     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
 829         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
 830     } else
 831         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
 832
 833     reversefill_audiodata(&s->silence, tmp_arg);
 834     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 835     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 836     return ret;
 837 }
 838
 839 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
 840     if (s->resampler && s->resample){
 841         return s->resampler->get_delay(s, base);
 842     }else{
 843         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
 844     }
 845 }
 846
 847 int swr_get_out_samples(struct SwrContext *s, int in_samples)
 848 {
 849     int64_t out_samples;
 850
 851     if (in_samples < 0)
 852         return AVERROR(EINVAL);
 853
 854     if (s->resampler && s->resample) {
 855         if (!s->resampler->get_out_samples)
 856             return AVERROR(ENOSYS);
 857         out_samples = s->resampler->get_out_samples(s, in_samples);
 858     } else {
 859         out_samples = s->in_buffer_count + in_samples;
 860         av_assert0(s->out_sample_rate == s->in_sample_rate);
 861     }
 862
 863     if (out_samples > INT_MAX)
 864         return AVERROR(EINVAL);
 865
 866     return out_samples;
 867 }
 868
 869 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
 870     int ret;
 871
 872     if (!s || compensation_distance < 0)
 873         return AVERROR(EINVAL);
 874     if (!compensation_distance && sample_delta)
 875         return AVERROR(EINVAL);
 876     if (!s->resample) {
 877         s->flags |= SWR_FLAG_RESAMPLE;
 878         ret = swr_init(s);
 879         if (ret < 0)
 880             return ret;
 881     }
 882     if (!s->resampler->set_compensation){
 883         return AVERROR(EINVAL);
 884     }else{
 885         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
 886     }
 887 }
 888
 889 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 890     if(pts == INT64_MIN)
 891         return s->outpts;
 892
 893     if (s->firstpts == AV_NOPTS_VALUE)
 894         s->outpts = s->firstpts = pts;
 895
 896     if(s->min_compensation >= FLT_MAX) {
 897         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 898     } else {
 899         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;
 900         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 901
 902         if(fabs(fdelta) > s->min_compensation) {
 903             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
 904                 int ret;
 905                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 906                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 907                 if(ret<0){
 908                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 909                 }
 910             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 911                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 912                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 913                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 914                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 915                 swr_set_compensation(s, comp, duration);
 916             }
 917         }
 918
 919         return s->outpts;
 920     }
 921 }