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/mem.h"
  22 #include "libavutil/opt.h"
  23 #include "swresample_internal.h"
  24 #include "audioconvert.h"
  25 #include "libavutil/avassert.h"
  26 #include "libavutil/channel_layout.h"
  27 #include "libavutil/internal.h"
  28
  29 #include <float.h>
  30
  31 #define ALIGN 32
  32
  33 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
  34     if(!s || s->in_convert) // s needs to be allocated but not initialized
  35         return AVERROR(EINVAL);
  36     s->channel_map = channel_map;
  37     return 0;
  38 }
  39
  40 int swr_alloc_set_opts2(struct SwrContext **ps,
  41                         const AVChannelLayout *out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  42                         const AVChannelLayout *in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
  43                         int log_offset, void *log_ctx) {
  44     struct SwrContext *s = *ps;
  45     int ret;
  46
  47     if (!s) s = swr_alloc();
  48     if (!s) return AVERROR(ENOMEM);
  49
  50     *ps = s;
  51
  52     s->log_level_offset = log_offset;
  53     s->log_ctx = log_ctx;
  54
  55     if ((ret = av_opt_set_chlayout(s, "ochl", out_ch_layout, 0)) < 0)
  56         goto fail;
  57
  58     if ((ret = av_opt_set_int(s, "osf", out_sample_fmt, 0)) < 0)
  59         goto fail;
  60
  61     if ((ret = av_opt_set_int(s, "osr", out_sample_rate, 0)) < 0)
  62         goto fail;
  63
  64     if ((ret = av_opt_set_chlayout(s, "ichl", in_ch_layout, 0)) < 0)
  65         goto fail;
  66
  67     if ((ret = av_opt_set_int(s, "isf", in_sample_fmt, 0)) < 0)
  68         goto fail;
  69
  70     if ((ret = av_opt_set_int(s, "isr", in_sample_rate, 0)) < 0)
  71         goto fail;
  72
  73     av_opt_set_int(s, "uch", 0, 0);
  74
  75     return 0;
  76 fail:
  77     av_log(s, AV_LOG_ERROR, "Failed to set option\n");
  78     swr_free(ps);
  79     return ret;
  80 }
  81
  82 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
  83     a->fmt   = fmt;
  84     a->bps   = av_get_bytes_per_sample(fmt);
  85     a->planar= av_sample_fmt_is_planar(fmt);
  86     if (a->ch_count == 1)
  87         a->planar = 1;
  88 }
  89
  90 static void free_temp(AudioData *a){
  91     av_free(a->data);
  92     memset(a, 0, sizeof(*a));
  93 }
  94
  95 static void clear_context(SwrContext *s){
  96     s->in_buffer_index= 0;
  97     s->in_buffer_count= 0;
  98     s->resample_in_constraint= 0;
  99     memset(s->in.ch, 0, sizeof(s->in.ch));
 100     memset(s->out.ch, 0, sizeof(s->out.ch));
 101     free_temp(&s->postin);
 102     free_temp(&s->midbuf);
 103     free_temp(&s->preout);
 104     free_temp(&s->in_buffer);
 105     free_temp(&s->silence);
 106     free_temp(&s->drop_temp);
 107     free_temp(&s->dither.noise);
 108     free_temp(&s->dither.temp);
 109     av_channel_layout_uninit(&s->in_ch_layout);
 110     av_channel_layout_uninit(&s->out_ch_layout);
 111     av_channel_layout_uninit(&s->used_ch_layout);
 112     swri_audio_convert_free(&s-> in_convert);
 113     swri_audio_convert_free(&s->out_convert);
 114     swri_audio_convert_free(&s->full_convert);
 115     swri_rematrix_free(s);
 116
 117     s->delayed_samples_fixup = 0;
 118     s->flushed = 0;
 119 }
 120
 121 av_cold void swr_free(SwrContext **ss){
 122     SwrContext *s= *ss;
 123     if(s){
 124         clear_context(s);
 125         av_channel_layout_uninit(&s->user_in_chlayout);
 126         av_channel_layout_uninit(&s->user_out_chlayout);
 127         av_channel_layout_uninit(&s->user_used_chlayout);
 128
 129         if (s->resampler)
 130             s->resampler->free(&s->resample);
 131     }
 132
 133     av_freep(ss);
 134 }
 135
 136 av_cold void swr_close(SwrContext *s){
 137     clear_context(s);
 138 }
 139
 140 av_cold int swr_init(struct SwrContext *s){
 141     int ret;
 142     char l1[1024], l2[1024];
 143
 144     clear_context(s);
 145
 146     if((unsigned) s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 147         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 148         return AVERROR(EINVAL);
 149     }
 150     if((unsigned) s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 151         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 152         return AVERROR(EINVAL);
 153     }
 154
 155     if(s-> in_sample_rate <= 0){
 156         av_log(s, AV_LOG_ERROR, "Requested input sample rate %d is invalid\n", s->in_sample_rate);
 157         return AVERROR(EINVAL);
 158     }
 159     if(s->out_sample_rate <= 0){
 160         av_log(s, AV_LOG_ERROR, "Requested output sample rate %d is invalid\n", s->out_sample_rate);
 161         return AVERROR(EINVAL);
 162     }
 163
 164     s->out.ch_count  = s-> user_out_chlayout.nb_channels;
 165     s-> in.ch_count  = s->  user_in_chlayout.nb_channels;
 166
 167     if (!(ret = av_channel_layout_check(&s->user_in_chlayout)) || s->user_in_chlayout.nb_channels > SWR_CH_MAX) {
 168         if (ret)
 169             av_channel_layout_describe(&s->user_in_chlayout, l1, sizeof(l1));
 170         av_log(s, AV_LOG_WARNING, "Input channel layout \"%s\" is invalid or unsupported.\n", ret ? l1 : "");
 171         return AVERROR(EINVAL);
 172     }
 173
 174     if (!(ret = av_channel_layout_check(&s->user_out_chlayout)) || s->user_out_chlayout.nb_channels > SWR_CH_MAX) {
 175         if (ret)
 176             av_channel_layout_describe(&s->user_out_chlayout, l2, sizeof(l2));
 177         av_log(s, AV_LOG_WARNING, "Output channel layout \"%s\" is invalid or unsupported.\n", ret ? l2 : "");
 178         return AVERROR(EINVAL);
 179     }
 180
 181     ret  = av_channel_layout_copy(&s->in_ch_layout, &s->user_in_chlayout);
 182     ret |= av_channel_layout_copy(&s->out_ch_layout, &s->user_out_chlayout);
 183     ret |= av_channel_layout_copy(&s->used_ch_layout, &s->user_used_chlayout);
 184     if (ret < 0)
 185         return ret;
 186
 187     s->int_sample_fmt= s->user_int_sample_fmt;
 188
 189     s->dither.method = s->user_dither_method;
 190
 191     switch(s->engine){
 192 #if CONFIG_LIBSOXR
 193         case SWR_ENGINE_SOXR: s->resampler = &swri_soxr_resampler; break;
 194 #endif
 195         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
 196         default:
 197             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
 198             return AVERROR(EINVAL);
 199     }
 200
 201     if (!av_channel_layout_check(&s->used_ch_layout))
 202         av_channel_layout_default(&s->used_ch_layout, s->in.ch_count);
 203
 204     if (s->used_ch_layout.nb_channels != s->in_ch_layout.nb_channels)
 205         av_channel_layout_uninit(&s->in_ch_layout);
 206
 207     if (s->used_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC)
 208         av_channel_layout_default(&s->used_ch_layout, s->used_ch_layout.nb_channels);
 209     if (s->in_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC) {
 210         ret = av_channel_layout_copy(&s->in_ch_layout, &s->used_ch_layout);
 211         if (ret < 0)
 212             return ret;
 213     }
 214     if (s->out_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC)
 215         av_channel_layout_default(&s->out_ch_layout, s->out.ch_count);
 216
 217     s->rematrix = av_channel_layout_compare(&s->out_ch_layout, &s->in_ch_layout) ||
 218                  s->rematrix_volume!=1.0 ||
 219                  s->rematrix_custom;
 220
 221     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 222         // 16bit or less to 16bit or less with the same sample rate
 223         if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
 224            && av_get_bytes_per_sample(s->out_sample_fmt) <= 2
 225            && s->out_sample_rate==s->in_sample_rate) {
 226             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 227         // 8 -> 8, 16->8, 8->16bit
 228         } else if(   av_get_bytes_per_sample(s-> in_sample_fmt)
 229                     +av_get_bytes_per_sample(s->out_sample_fmt) <= 3 ) {
 230             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 231         }else if(   av_get_bytes_per_sample(s-> in_sample_fmt) <= 2
 232            && !s->rematrix
 233            && s->out_sample_rate==s->in_sample_rate
 234            && !(s->flags & SWR_FLAG_RESAMPLE)){
 235             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 236         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
 237                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
 238                  && !s->rematrix
 239                  && s->out_sample_rate == s->in_sample_rate
 240                  && !(s->flags & SWR_FLAG_RESAMPLE)
 241                  && s->engine != SWR_ENGINE_SOXR){
 242             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
 243         }else if(av_get_bytes_per_sample(s->in_sample_fmt) <= 4){
 244             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 245         }else{
 246             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 247         }
 248     }
 249     av_log(s, AV_LOG_DEBUG, "Using %s internally between filters\n", av_get_sample_fmt_name(s->int_sample_fmt));
 250
 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_S64P
 254         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 255         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 256         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, s16p/s32p/s64p/fltp/dblp are supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
 257         return AVERROR(EINVAL);
 258     }
 259
 260     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 261     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 262
 263     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
 264         if (!s->async && s->min_compensation >= FLT_MAX/2)
 265             s->async = 1;
 266         if (s->firstpts == AV_NOPTS_VALUE)
 267             s->firstpts =
 268             s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
 269     } else
 270         s->firstpts = AV_NOPTS_VALUE;
 271
 272     if (s->async) {
 273         if (s->min_compensation >= FLT_MAX/2)
 274             s->min_compensation = 0.001;
 275         if (s->async > 1.0001) {
 276             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
 277         }
 278     }
 279
 280     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 281         s->resample = s->resampler->init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta, s->precision, s->cheby, s->exact_rational);
 282         if (!s->resample) {
 283             av_log(s, AV_LOG_ERROR, "Failed to initialize resampler\n");
 284             return AVERROR(ENOMEM);
 285         }
 286     }else
 287         s->resampler->free(&s->resample);
 288     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 289         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 290         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 291         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 292         && s->resample){
 293         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16p/s32p/fltp/dblp\n");
 294         ret = AVERROR(EINVAL);
 295         goto fail;
 296     }
 297
 298 #define RSC 1 //FIXME finetune
 299     if(!s-> in.ch_count)
 300         s-> in.ch_count = s->in_ch_layout.nb_channels;
 301     if (!av_channel_layout_check(&s->used_ch_layout))
 302         av_channel_layout_default(&s->used_ch_layout, s->in.ch_count);
 303     if(!s->out.ch_count)
 304         s->out.ch_count = s->out_ch_layout.nb_channels;
 305
 306     if(!s-> in.ch_count){
 307         av_assert0(s->in_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC);
 308         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 309         ret = AVERROR(EINVAL);
 310         goto fail;
 311     }
 312
 313     av_channel_layout_describe(&s->out_ch_layout, l2, sizeof(l2));
 314     av_channel_layout_describe(&s->in_ch_layout, l1, sizeof(l1));
 315     if (s->in_ch_layout.order != AV_CHANNEL_ORDER_UNSPEC && s->used_ch_layout.nb_channels != s->in_ch_layout.nb_channels) {
 316         av_log(s, AV_LOG_ERROR, "Input channel layout %s mismatches specified channel count %d\n", l1, s->used_ch_layout.nb_channels);
 317         ret = AVERROR(EINVAL);
 318         goto fail;
 319     }
 320
 321     if ((   s->out_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC
 322          || s-> in_ch_layout.order == AV_CHANNEL_ORDER_UNSPEC) && s->used_ch_layout.nb_channels != s->out.ch_count && !s->rematrix_custom) {
 323         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
 324                "but there is not enough information to do it\n", l1, l2);
 325         ret = AVERROR(EINVAL);
 326         goto fail;
 327     }
 328
 329 av_assert0(s->used_ch_layout.nb_channels);
 330 av_assert0(s->out.ch_count);
 331     s->resample_first= RSC*s->out.ch_count/s->used_ch_layout.nb_channels - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 332
 333     s->in_buffer= s->in;
 334     s->silence  = s->in;
 335     s->drop_temp= s->out;
 336
 337     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
 338         goto fail;
 339
 340     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
 341         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 342                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 343         return 0;
 344     }
 345
 346     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 347                                              s-> in_sample_fmt, s->used_ch_layout.nb_channels, s->channel_map, 0);
 348     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 349                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 350
 351     if (!s->in_convert || !s->out_convert) {
 352         ret = AVERROR(ENOMEM);
 353         goto fail;
 354     }
 355
 356     s->postin= s->in;
 357     s->preout= s->out;
 358     s->midbuf= s->in;
 359
 360     if(s->channel_map){
 361         s->postin.ch_count=
 362         s->midbuf.ch_count= s->used_ch_layout.nb_channels;
 363         if(s->resample)
 364             s->in_buffer.ch_count= s->used_ch_layout.nb_channels;
 365     }
 366     if(!s->resample_first){
 367         s->midbuf.ch_count= s->out.ch_count;
 368         if(s->resample)
 369             s->in_buffer.ch_count = s->out.ch_count;
 370     }
 371
 372     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 373     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 374     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 375
 376     if(s->resample){
 377         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 378     }
 379
 380     av_assert0(!s->preout.count);
 381     s->dither.noise = s->preout;
 382     s->dither.temp  = s->preout;
 383     if (s->dither.method > SWR_DITHER_NS) {
 384         s->dither.noise.bps = 4;
 385         s->dither.noise.fmt = AV_SAMPLE_FMT_FLTP;
 386         s->dither.noise_scale = 1;
 387     }
 388
 389     if(s->rematrix || s->dither.method) {
 390         ret = swri_rematrix_init(s);
 391         if (ret < 0)
 392             goto fail;
 393     }
 394
 395     return 0;
 396 fail:
 397     swr_close(s);
 398     return ret;
 399
 400 }
 401
 402 int swri_realloc_audio(AudioData *a, int count){
 403     int i, countb;
 404     AudioData old;
 405
 406     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 407         return AVERROR(EINVAL);
 408
 409     if(a->count >= count)
 410         return 0;
 411
 412     count*=2;
 413
 414     countb= FFALIGN(count*a->bps, ALIGN);
 415     old= *a;
 416
 417     av_assert0(a->bps);
 418     av_assert0(a->ch_count);
 419
 420     a->data = av_calloc(countb, a->ch_count);
 421     if(!a->data)
 422         return AVERROR(ENOMEM);
 423     for(i=0; i<a->ch_count; i++){
 424         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 425         if(a->count && a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 426     }
 427     if(a->count && !a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 428     av_freep(&old.data);
 429     a->count= count;
 430
 431     return 1;
 432 }
 433
 434 static void copy(AudioData *out, AudioData *in,
 435                  int count){
 436     av_assert0(out->planar == in->planar);
 437     av_assert0(out->bps == in->bps);
 438     av_assert0(out->ch_count == in->ch_count);
 439     if(out->planar){
 440         int ch;
 441         for(ch=0; ch<out->ch_count; ch++)
 442             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 443     }else
 444         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 445 }
 446
 447 static void fill_audiodata(AudioData *out, uint8_t *const in_arg [SWR_CH_MAX])
 448 {
 449     int i;
 450     if(!in_arg){
 451         memset(out->ch, 0, sizeof(out->ch));
 452     }else if(out->planar){
 453         for(i=0; i<out->ch_count; i++)
 454             out->ch[i]= in_arg[i];
 455     }else{
 456         for(i=0; i<out->ch_count; i++)
 457             out->ch[i]= in_arg[0] + i*out->bps;
 458     }
 459 }
 460
 461 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 462     int i;
 463     if(out->planar){
 464         for(i=0; i<out->ch_count; i++)
 465             in_arg[i]= out->ch[i];
 466     }else{
 467         in_arg[0]= out->ch[0];
 468     }
 469 }
 470
 471 /**
 472  *
 473  * out may be equal in.
 474  */
 475 static void buf_set(AudioData *out, AudioData *in, int count){
 476     int ch;
 477     if(in->planar){
 478         for(ch=0; ch<out->ch_count; ch++)
 479             out->ch[ch]= in->ch[ch] + count*out->bps;
 480     }else{
 481         for(ch=out->ch_count-1; ch>=0; ch--)
 482             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 483     }
 484 }
 485
 486 /**
 487  *
 488  * @return number of samples output per channel
 489  */
 490 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 491                              const AudioData * in_param, int in_count){
 492     AudioData in, out, tmp;
 493     int ret_sum=0;
 494     int border=0;
 495     int padless = ARCH_X86 && s->engine == SWR_ENGINE_SWR ? 7 : 0;
 496
 497     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 498     av_assert1(s->in_buffer.planar   == in_param->planar);
 499     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 500
 501     tmp=out=*out_param;
 502     in =  *in_param;
 503
 504     border = s->resampler->invert_initial_buffer(s->resample, &s->in_buffer,
 505                  &in, in_count, &s->in_buffer_index, &s->in_buffer_count);
 506     if (border == INT_MAX) {
 507         return 0;
 508     } else if (border < 0) {
 509         return border;
 510     } else if (border) {
 511         buf_set(&in, &in, border);
 512         in_count -= border;
 513         s->resample_in_constraint = 0;
 514     }
 515
 516     do{
 517         int ret, size, consumed;
 518         if(!s->resample_in_constraint && s->in_buffer_count){
 519             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 520             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 521             out_count -= ret;
 522             ret_sum += ret;
 523             buf_set(&out, &out, ret);
 524             s->in_buffer_count -= consumed;
 525             s->in_buffer_index += consumed;
 526
 527             if(!in_count)
 528                 break;
 529             if(s->in_buffer_count <= border){
 530                 buf_set(&in, &in, -s->in_buffer_count);
 531                 in_count += s->in_buffer_count;
 532                 s->in_buffer_count=0;
 533                 s->in_buffer_index=0;
 534                 border = 0;
 535             }
 536         }
 537
 538         if((s->flushed || in_count > padless) && !s->in_buffer_count){
 539             s->in_buffer_index=0;
 540             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, FFMAX(in_count-padless, 0), &consumed);
 541             out_count -= ret;
 542             ret_sum += ret;
 543             buf_set(&out, &out, ret);
 544             in_count -= consumed;
 545             buf_set(&in, &in, consumed);
 546         }
 547
 548         //TODO is this check sane considering the advanced copy avoidance below
 549         size= s->in_buffer_index + s->in_buffer_count + in_count;
 550         if(   size > s->in_buffer.count
 551            && s->in_buffer_count + in_count <= s->in_buffer_index){
 552             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 553             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 554             s->in_buffer_index=0;
 555         }else
 556             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 557                 return ret;
 558
 559         if(in_count){
 560             int count= in_count;
 561             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 562
 563             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 564             copy(&tmp, &in, /*in_*/count);
 565             s->in_buffer_count += count;
 566             in_count -= count;
 567             border += count;
 568             buf_set(&in, &in, count);
 569             s->resample_in_constraint= 0;
 570             if(s->in_buffer_count != count || in_count)
 571                 continue;
 572             if (padless) {
 573                 padless = 0;
 574                 continue;
 575             }
 576         }
 577         break;
 578     }while(1);
 579
 580     s->resample_in_constraint= !!out_count;
 581
 582     return ret_sum;
 583 }
 584
 585 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 586                                                       AudioData *in , int  in_count){
 587     AudioData *postin, *midbuf, *preout;
 588     int ret/*, in_max*/;
 589     AudioData preout_tmp, midbuf_tmp;
 590
 591     if(s->full_convert){
 592         av_assert0(!s->resample);
 593         swri_audio_convert(s->full_convert, out, in, in_count);
 594         return out_count;
 595     }
 596
 597 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 598 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 599
 600     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
 601         return ret;
 602     if(s->resample_first){
 603         av_assert0(s->midbuf.ch_count == s->used_ch_layout.nb_channels);
 604         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
 605             return ret;
 606     }else{
 607         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 608         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
 609             return ret;
 610     }
 611     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
 612         return ret;
 613
 614     postin= &s->postin;
 615
 616     midbuf_tmp= s->midbuf;
 617     midbuf= &midbuf_tmp;
 618     preout_tmp= s->preout;
 619     preout= &preout_tmp;
 620
 621     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 622         postin= in;
 623
 624     if(s->resample_first ? !s->resample : !s->rematrix)
 625         midbuf= postin;
 626
 627     if(s->resample_first ? !s->rematrix : !s->resample)
 628         preout= midbuf;
 629
 630     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar
 631        && !(s->out_sample_fmt==AV_SAMPLE_FMT_S32P && (s->dither.output_sample_bits&31))){
 632         if(preout==in){
 633             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 634             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 635             copy(out, in, out_count);
 636             return out_count;
 637         }
 638         else if(preout==postin) preout= midbuf= postin= out;
 639         else if(preout==midbuf) preout= midbuf= out;
 640         else                    preout= out;
 641     }
 642
 643     if(in != postin){
 644         swri_audio_convert(s->in_convert, postin, in, in_count);
 645     }
 646
 647     if(s->resample_first){
 648         if(postin != midbuf)
 649             if ((out_count = resample(s, midbuf, out_count, postin, in_count)) < 0)
 650                 return out_count;
 651         if(midbuf != preout)
 652             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 653     }else{
 654         if(postin != midbuf)
 655             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 656         if(midbuf != preout)
 657             if ((out_count = resample(s, preout, out_count, midbuf, in_count)) < 0)
 658                 return out_count;
 659     }
 660
 661     if(preout != out && out_count){
 662         AudioData *conv_src = preout;
 663         if(s->dither.method){
 664             int ch;
 665             int dither_count= FFMAX(out_count, 1<<16);
 666
 667             if (preout == in) {
 668                 conv_src = &s->dither.temp;
 669                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
 670                     return ret;
 671             }
 672
 673             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
 674                 return ret;
 675             if(ret)
 676                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
 677                     if((ret=swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, (12345678913579ULL*ch + 3141592) % 2718281828U, s->dither.noise.fmt))<0)
 678                         return ret;
 679             av_assert0(s->dither.noise.ch_count == preout->ch_count);
 680
 681             if(s->dither.noise_pos + out_count > s->dither.noise.count)
 682                 s->dither.noise_pos = 0;
 683
 684             if (s->dither.method < SWR_DITHER_NS){
 685                 if (s->mix_2_1_simd) {
 686                     int len1= out_count&~15;
 687                     int off = len1 * preout->bps;
 688
 689                     if(len1)
 690                         for(ch=0; ch<preout->ch_count; ch++)
 691                             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);
 692                     if(out_count != len1)
 693                         for(ch=0; ch<preout->ch_count; ch++)
 694                             s->mix_2_1_f(conv_src->ch[ch] + off, preout->ch[ch] + off, s->dither.noise.ch[ch] + s->dither.noise.bps * s->dither.noise_pos + off, s->native_one, 0, 0, out_count - len1);
 695                 } else {
 696                     for(ch=0; ch<preout->ch_count; ch++)
 697                         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);
 698                 }
 699             } else {
 700                 switch(s->int_sample_fmt) {
 701                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
 702                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
 703                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
 704                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
 705                 }
 706             }
 707             s->dither.noise_pos += out_count;
 708         }
 709 //FIXME packed doesn't need more than 1 chan here!
 710         swri_audio_convert(s->out_convert, out, conv_src, out_count);
 711     }
 712     return out_count;
 713 }
 714
 715 int swr_is_initialized(struct SwrContext *s) {
 716     return !!s->in_buffer.ch_count;
 717 }
 718
 719 int attribute_align_arg swr_convert(struct SwrContext *s,
 720                                           uint8_t * const *out_arg, int out_count,
 721                                     const uint8_t * const *in_arg,  int in_count)
 722 {
 723     AudioData * in= &s->in;
 724     AudioData *out= &s->out;
 725     int av_unused max_output;
 726
 727     if (!swr_is_initialized(s)) {
 728         av_log(s, AV_LOG_ERROR, "Context has not been initialized\n");
 729         return AVERROR(EINVAL);
 730     }
 731 #if defined(ASSERT_LEVEL) && ASSERT_LEVEL >1
 732     max_output = swr_get_out_samples(s, in_count);
 733 #endif
 734
 735     while(s->drop_output > 0){
 736         int ret;
 737         uint8_t *tmp_arg[SWR_CH_MAX];
 738 #define MAX_DROP_STEP 16384
 739         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
 740             return ret;
 741
 742         reversefill_audiodata(&s->drop_temp, tmp_arg);
 743         s->drop_output *= -1; //FIXME find a less hackish solution
 744         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
 745         s->drop_output *= -1;
 746         in_count = 0;
 747         if(ret>0) {
 748             s->drop_output -= ret;
 749             if (!s->drop_output && !out_arg)
 750                 return 0;
 751             continue;
 752         }
 753
 754         av_assert0(s->drop_output);
 755         return 0;
 756     }
 757
 758     if(!in_arg){
 759         if(s->resample){
 760             if (!s->flushed)
 761                 s->resampler->flush(s);
 762             s->resample_in_constraint = 0;
 763             s->flushed = 1;
 764         }else if(!s->in_buffer_count){
 765             return 0;
 766         }
 767     }else
 768         fill_audiodata(in ,  (void*)in_arg);
 769
 770     fill_audiodata(out, out_arg);
 771
 772     if(s->resample){
 773         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 774         if(ret>0 && !s->drop_output)
 775             s->outpts += ret * (int64_t)s->in_sample_rate;
 776
 777         av_assert2(max_output < 0 || ret <= max_output);
 778
 779         return ret;
 780     }else{
 781         AudioData tmp= *in;
 782         int ret2=0;
 783         int ret, size;
 784         size = FFMIN(out_count, s->in_buffer_count);
 785         if(size){
 786             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 787             ret= swr_convert_internal(s, out, size, &tmp, size);
 788             if(ret<0)
 789                 return ret;
 790             ret2= ret;
 791             s->in_buffer_count -= ret;
 792             s->in_buffer_index += ret;
 793             buf_set(out, out, ret);
 794             out_count -= ret;
 795             if(!s->in_buffer_count)
 796                 s->in_buffer_index = 0;
 797         }
 798
 799         if(in_count){
 800             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 801
 802             if(in_count > out_count) { //FIXME move after swr_convert_internal
 803                 if(   size > s->in_buffer.count
 804                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 805                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 806                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 807                     s->in_buffer_index=0;
 808                 }else
 809                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 810                         return ret;
 811             }
 812
 813             if(out_count){
 814                 size = FFMIN(in_count, out_count);
 815                 ret= swr_convert_internal(s, out, size, in, size);
 816                 if(ret<0)
 817                     return ret;
 818                 buf_set(in, in, ret);
 819                 in_count -= ret;
 820                 ret2 += ret;
 821             }
 822             if(in_count){
 823                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 824                 copy(&tmp, in, in_count);
 825                 s->in_buffer_count += in_count;
 826             }
 827         }
 828         if(ret2>0 && !s->drop_output)
 829             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 830         av_assert2(max_output < 0 || ret2 < 0 || ret2 <= max_output);
 831         return ret2;
 832     }
 833 }
 834
 835 int swr_drop_output(struct SwrContext *s, int count){
 836     const uint8_t *tmp_arg[SWR_CH_MAX];
 837     s->drop_output += count;
 838
 839     if(s->drop_output <= 0)
 840         return 0;
 841
 842     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 843     return swr_convert(s, NULL, s->drop_output, tmp_arg, 0);
 844 }
 845
 846 int swr_inject_silence(struct SwrContext *s, int count){
 847     int ret, i;
 848     uint8_t *tmp_arg[SWR_CH_MAX];
 849
 850     if(count <= 0)
 851         return 0;
 852
 853 #define MAX_SILENCE_STEP 16384
 854     while (count > MAX_SILENCE_STEP) {
 855         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
 856             return ret;
 857         count -= MAX_SILENCE_STEP;
 858     }
 859
 860     if((ret=swri_realloc_audio(&s->silence, count))<0)
 861         return ret;
 862
 863     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
 864         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
 865     } else
 866         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
 867
 868     reversefill_audiodata(&s->silence, tmp_arg);
 869     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 870     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 871     return ret;
 872 }
 873
 874 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
 875     if (s->resampler && s->resample){
 876         return s->resampler->get_delay(s, base);
 877     }else{
 878         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
 879     }
 880 }
 881
 882 int swr_get_out_samples(struct SwrContext *s, int in_samples)
 883 {
 884     int64_t out_samples;
 885
 886     if (in_samples < 0)
 887         return AVERROR(EINVAL);
 888
 889     if (s->resampler && s->resample) {
 890         if (!s->resampler->get_out_samples)
 891             return AVERROR(ENOSYS);
 892         out_samples = s->resampler->get_out_samples(s, in_samples);
 893     } else {
 894         out_samples = s->in_buffer_count + in_samples;
 895         av_assert0(s->out_sample_rate == s->in_sample_rate);
 896     }
 897
 898     if (out_samples > INT_MAX)
 899         return AVERROR(EINVAL);
 900
 901     return out_samples;
 902 }
 903
 904 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
 905     int ret;
 906
 907     if (!s || compensation_distance < 0)
 908         return AVERROR(EINVAL);
 909     if (!compensation_distance && sample_delta)
 910         return AVERROR(EINVAL);
 911     if (!s->resample) {
 912         s->flags |= SWR_FLAG_RESAMPLE;
 913         ret = swr_init(s);
 914         if (ret < 0)
 915             return ret;
 916     }
 917     if (!s->resampler->set_compensation){
 918         return AVERROR(EINVAL);
 919     }else{
 920         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
 921     }
 922 }
 923
 924 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 925     if(pts == INT64_MIN)
 926         return s->outpts;
 927
 928     if (s->firstpts == AV_NOPTS_VALUE)
 929         s->outpts = s->firstpts = pts;
 930
 931     if(s->min_compensation >= FLT_MAX) {
 932         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 933     } else {
 934         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;
 935         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 936
 937         if(fabs(fdelta) > s->min_compensation) {
 938             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
 939                 int ret;
 940                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 941                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 942                 if(ret<0){
 943                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 944                 }
 945             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 946                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 947                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 948                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 949                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 950                 swr_set_compensation(s, comp, duration);
 951             }
 952         }
 953
 954         return s->outpts;
 955     }
 956 }