git.sesse.net Git - ffmpeg/blob - libswresample/swresample.c

   1 /*
   2  * Copyright (C) 2011-2012 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/audioconvert.h"
  26
  27 #include <float.h>
  28
  29 #define  C30DB  M_SQRT2
  30 #define  C15DB  1.189207115
  31 #define C__0DB  1.0
  32 #define C_15DB  0.840896415
  33 #define C_30DB  M_SQRT1_2
  34 #define C_45DB  0.594603558
  35 #define C_60DB  0.5
  36
  37 #define ALIGN 32
  38
  39 //TODO split options array out?
  40 #define OFFSET(x) offsetof(SwrContext,x)
  41 #define PARAM AV_OPT_FLAG_AUDIO_PARAM
  42
  43 static const AVOption options[]={
  44 {"ich"                  ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
  45 {"in_channel_count"     ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
  46 {"och"                  , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
  47 {"out_channel_count"    , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.dbl=2                     }, 0      , SWR_CH_MAX, PARAM},
  48 {"uch"                  ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , SWR_CH_MAX, PARAM},
  49 {"used_channel_count"   ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , SWR_CH_MAX, PARAM},
  50 {"isr"                  ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
  51 {"in_sample_rate"       ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
  52 {"osr"                  , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
  53 {"out_sample_rate"      , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , INT_MAX   , PARAM},
  54 {"isf"                  ,    "Input Sample Format"      , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  55 {"in_sample_fmt"        ,    "Input Sample Format"      , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  56 {"osf"                  ,   "Output Sample Format"      , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  57 {"out_sample_fmt"       ,   "Output Sample Format"      , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  58 {"tsf"                  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
  59 {"internal_sample_fmt"  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.dbl=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
  60 {"icl"                  ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  61 {"in_channel_layout"    ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  62 {"ocl"                  ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  63 {"out_channel_layout"   ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.dbl=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  64 {"clev"                 ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  65 {"center_mix_level"     ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  66 {"slev"                 , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  67 {"surround_mix_level"   , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  68 {"lfe_mix_level"        , "LFE Mix Level"               , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
  69 {"rmvol"                , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
  70 {"rematrix_volume"      , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
  71 {"flags"                , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.dbl=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
  72 {"swr_flags"            , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.dbl=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
  73 {"res"                  , "Force Resampling"            , 0                      , AV_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
  74 {"dither_scale"         , "Dither Scale"                , OFFSET(dither_scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
  75 {"dither_method"        , "Dither Method"               , OFFSET(dither_method  ), AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
  76 {"rectangular"          , "Rectangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
  77 {"triangular"           ,  "Triangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
  78 {"triangular_hp"        , "Triangular Dither With High Pass" , 0                 , AV_OPT_TYPE_CONST, {.dbl=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  79 {"filter_size"          , "Resampling Filter Size"      , OFFSET(filter_size)    , AV_OPT_TYPE_INT  , {.dbl=16                    }, 0      , INT_MAX   , PARAM },
  80 {"phase_shift"          , "Resampling Phase Shift"      , OFFSET(phase_shift)    , AV_OPT_TYPE_INT  , {.dbl=10                    }, 0      , 30        , PARAM },
  81 {"linear_interp"        , "Use Linear Interpolation"    , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.dbl=0                     }, 0      , 1         , PARAM },
  82 {"cutoff"               , "Cutoff Frequency Ratio"      , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.8                   }, 0      , 1         , PARAM },
  83 {"min_comp"             , "Minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
  84                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
  85 {"min_hard_comp"        , "Minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
  86                                                    , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
  87 {"comp_duration"        , "Duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
  88                                               , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
  89 {"max_soft_comp"        , "Maximum factor by which data is stretched/squeezed to make it match the timestamps."
  90                                                    , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
  91
  92 {0}
  93 };
  94
  95 static const char* context_to_name(void* ptr) {
  96     return "SWR";
  97 }
  98
  99 static const AVClass av_class = {
 100     .class_name                = "SwrContext",
 101     .item_name                 = context_to_name,
 102     .option                    = options,
 103     .version                   = LIBAVUTIL_VERSION_INT,
 104     .log_level_offset_offset   = OFFSET(log_level_offset),
 105     .parent_log_context_offset = OFFSET(log_ctx),
 106 };
 107
 108 unsigned swresample_version(void)
 109 {
 110     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
 111     return LIBSWRESAMPLE_VERSION_INT;
 112 }
 113
 114 const char *swresample_configuration(void)
 115 {
 116     return FFMPEG_CONFIGURATION;
 117 }
 118
 119 const char *swresample_license(void)
 120 {
 121 #define LICENSE_PREFIX "libswresample license: "
 122     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
 123 }
 124
 125 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
 126     if(!s || s->in_convert) // s needs to be allocated but not initialized
 127         return AVERROR(EINVAL);
 128     s->channel_map = channel_map;
 129     return 0;
 130 }
 131
 132 const AVClass *swr_get_class(void)
 133 {
 134     return &av_class;
 135 }
 136
 137 struct SwrContext *swr_alloc(void){
 138     SwrContext *s= av_mallocz(sizeof(SwrContext));
 139     if(s){
 140         s->av_class= &av_class;
 141         av_opt_set_defaults(s);
 142     }
 143     return s;
 144 }
 145
 146 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
 147                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
 148                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
 149                                       int log_offset, void *log_ctx){
 150     if(!s) s= swr_alloc();
 151     if(!s) return NULL;
 152
 153     s->log_level_offset= log_offset;
 154     s->log_ctx= log_ctx;
 155
 156     av_opt_set_int(s, "ocl", out_ch_layout,   0);
 157     av_opt_set_int(s, "osf", out_sample_fmt,  0);
 158     av_opt_set_int(s, "osr", out_sample_rate, 0);
 159     av_opt_set_int(s, "icl", in_ch_layout,    0);
 160     av_opt_set_int(s, "isf", in_sample_fmt,   0);
 161     av_opt_set_int(s, "isr", in_sample_rate,  0);
 162     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
 163     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
 164     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
 165     av_opt_set_int(s, "uch", 0, 0);
 166     return s;
 167 }
 168
 169 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 170     a->fmt   = fmt;
 171     a->bps   = av_get_bytes_per_sample(fmt);
 172     a->planar= av_sample_fmt_is_planar(fmt);
 173 }
 174
 175 static void free_temp(AudioData *a){
 176     av_free(a->data);
 177     memset(a, 0, sizeof(*a));
 178 }
 179
 180 void swr_free(SwrContext **ss){
 181     SwrContext *s= *ss;
 182     if(s){
 183         free_temp(&s->postin);
 184         free_temp(&s->midbuf);
 185         free_temp(&s->preout);
 186         free_temp(&s->in_buffer);
 187         free_temp(&s->dither);
 188         swri_audio_convert_free(&s-> in_convert);
 189         swri_audio_convert_free(&s->out_convert);
 190         swri_audio_convert_free(&s->full_convert);
 191         swri_resample_free(&s->resample);
 192         swri_rematrix_free(s);
 193     }
 194
 195     av_freep(ss);
 196 }
 197
 198 int swr_init(struct SwrContext *s){
 199     s->in_buffer_index= 0;
 200     s->in_buffer_count= 0;
 201     s->resample_in_constraint= 0;
 202     free_temp(&s->postin);
 203     free_temp(&s->midbuf);
 204     free_temp(&s->preout);
 205     free_temp(&s->in_buffer);
 206     free_temp(&s->dither);
 207     swri_audio_convert_free(&s-> in_convert);
 208     swri_audio_convert_free(&s->out_convert);
 209     swri_audio_convert_free(&s->full_convert);
 210     swri_rematrix_free(s);
 211
 212     s->flushed = 0;
 213
 214     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 215         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 216         return AVERROR(EINVAL);
 217     }
 218     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 219         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 220         return AVERROR(EINVAL);
 221     }
 222
 223     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 224         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 225             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 226         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 227             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 228         }else{
 229             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 230             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 231         }
 232     }
 233
 234     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 235         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 236         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 237         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 238         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));
 239         return AVERROR(EINVAL);
 240     }
 241
 242     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 243     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 244
 245     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 246         s->resample = swri_resample_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);
 247     }else
 248         swri_resample_free(&s->resample);
 249     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 250         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 251         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 252         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 253         && s->resample){
 254         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 255         return -1;
 256     }
 257
 258     if(!s->used_ch_count)
 259         s->used_ch_count= s->in.ch_count;
 260
 261     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 262         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 263         s-> in_ch_layout= 0;
 264     }
 265
 266     if(!s-> in_ch_layout)
 267         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 268     if(!s->out_ch_layout)
 269         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 270
 271     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 272                  s->rematrix_custom;
 273
 274 #define RSC 1 //FIXME finetune
 275     if(!s-> in.ch_count)
 276         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 277     if(!s->used_ch_count)
 278         s->used_ch_count= s->in.ch_count;
 279     if(!s->out.ch_count)
 280         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 281
 282     if(!s-> in.ch_count){
 283         av_assert0(!s->in_ch_layout);
 284         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 285         return -1;
 286     }
 287
 288     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 289         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
 290         return -1;
 291     }
 292
 293 av_assert0(s->used_ch_count);
 294 av_assert0(s->out.ch_count);
 295     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 296
 297     s->in_buffer= s->in;
 298
 299     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
 300         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 301                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 302         return 0;
 303     }
 304
 305     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 306                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 307     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 308                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 309
 310
 311     s->postin= s->in;
 312     s->preout= s->out;
 313     s->midbuf= s->in;
 314
 315     if(s->channel_map){
 316         s->postin.ch_count=
 317         s->midbuf.ch_count= s->used_ch_count;
 318         if(s->resample)
 319             s->in_buffer.ch_count= s->used_ch_count;
 320     }
 321     if(!s->resample_first){
 322         s->midbuf.ch_count= s->out.ch_count;
 323         if(s->resample)
 324             s->in_buffer.ch_count = s->out.ch_count;
 325     }
 326
 327     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 328     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 329     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 330
 331     if(s->resample){
 332         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 333     }
 334
 335     s->dither = s->preout;
 336
 337     if(s->rematrix || s->dither_method)
 338         return swri_rematrix_init(s);
 339
 340     return 0;
 341 }
 342
 343 static int realloc_audio(AudioData *a, int count){
 344     int i, countb;
 345     AudioData old;
 346
 347     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 348         return AVERROR(EINVAL);
 349
 350     if(a->count >= count)
 351         return 0;
 352
 353     count*=2;
 354
 355     countb= FFALIGN(count*a->bps, ALIGN);
 356     old= *a;
 357
 358     av_assert0(a->bps);
 359     av_assert0(a->ch_count);
 360
 361     a->data= av_mallocz(countb*a->ch_count);
 362     if(!a->data)
 363         return AVERROR(ENOMEM);
 364     for(i=0; i<a->ch_count; i++){
 365         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 366         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 367     }
 368     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 369     av_free(old.data);
 370     a->count= count;
 371
 372     return 1;
 373 }
 374
 375 static void copy(AudioData *out, AudioData *in,
 376                  int count){
 377     av_assert0(out->planar == in->planar);
 378     av_assert0(out->bps == in->bps);
 379     av_assert0(out->ch_count == in->ch_count);
 380     if(out->planar){
 381         int ch;
 382         for(ch=0; ch<out->ch_count; ch++)
 383             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 384     }else
 385         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 386 }
 387
 388 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 389     int i;
 390     if(!in_arg){
 391         memset(out->ch, 0, sizeof(out->ch));
 392     }else if(out->planar){
 393         for(i=0; i<out->ch_count; i++)
 394             out->ch[i]= in_arg[i];
 395     }else{
 396         for(i=0; i<out->ch_count; i++)
 397             out->ch[i]= in_arg[0] + i*out->bps;
 398     }
 399 }
 400
 401 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 402     int i;
 403     if(out->planar){
 404         for(i=0; i<out->ch_count; i++)
 405             in_arg[i]= out->ch[i];
 406     }else{
 407         in_arg[0]= out->ch[0];
 408     }
 409 }
 410
 411 /**
 412  *
 413  * out may be equal in.
 414  */
 415 static void buf_set(AudioData *out, AudioData *in, int count){
 416     int ch;
 417     if(in->planar){
 418         for(ch=0; ch<out->ch_count; ch++)
 419             out->ch[ch]= in->ch[ch] + count*out->bps;
 420     }else{
 421         for(ch=out->ch_count-1; ch>=0; ch--)
 422             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 423     }
 424 }
 425
 426 /**
 427  *
 428  * @return number of samples output per channel
 429  */
 430 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 431                              const AudioData * in_param, int in_count){
 432     AudioData in, out, tmp;
 433     int ret_sum=0;
 434     int border=0;
 435
 436     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 437     av_assert1(s->in_buffer.planar   == in_param->planar);
 438     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 439
 440     tmp=out=*out_param;
 441     in =  *in_param;
 442
 443     do{
 444         int ret, size, consumed;
 445         if(!s->resample_in_constraint && s->in_buffer_count){
 446             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 447             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 448             out_count -= ret;
 449             ret_sum += ret;
 450             buf_set(&out, &out, ret);
 451             s->in_buffer_count -= consumed;
 452             s->in_buffer_index += consumed;
 453
 454             if(!in_count)
 455                 break;
 456             if(s->in_buffer_count <= border){
 457                 buf_set(&in, &in, -s->in_buffer_count);
 458                 in_count += s->in_buffer_count;
 459                 s->in_buffer_count=0;
 460                 s->in_buffer_index=0;
 461                 border = 0;
 462             }
 463         }
 464
 465         if(in_count && !s->in_buffer_count){
 466             s->in_buffer_index=0;
 467             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 468             out_count -= ret;
 469             ret_sum += ret;
 470             buf_set(&out, &out, ret);
 471             in_count -= consumed;
 472             buf_set(&in, &in, consumed);
 473         }
 474
 475         //TODO is this check sane considering the advanced copy avoidance below
 476         size= s->in_buffer_index + s->in_buffer_count + in_count;
 477         if(   size > s->in_buffer.count
 478            && s->in_buffer_count + in_count <= s->in_buffer_index){
 479             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 480             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 481             s->in_buffer_index=0;
 482         }else
 483             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 484                 return ret;
 485
 486         if(in_count){
 487             int count= in_count;
 488             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 489
 490             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 491             copy(&tmp, &in, /*in_*/count);
 492             s->in_buffer_count += count;
 493             in_count -= count;
 494             border += count;
 495             buf_set(&in, &in, count);
 496             s->resample_in_constraint= 0;
 497             if(s->in_buffer_count != count || in_count)
 498                 continue;
 499         }
 500         break;
 501     }while(1);
 502
 503     s->resample_in_constraint= !!out_count;
 504
 505     return ret_sum;
 506 }
 507
 508 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 509                                                       AudioData *in , int  in_count){
 510     AudioData *postin, *midbuf, *preout;
 511     int ret/*, in_max*/;
 512     AudioData preout_tmp, midbuf_tmp;
 513
 514     if(s->full_convert){
 515         av_assert0(!s->resample);
 516         swri_audio_convert(s->full_convert, out, in, in_count);
 517         return out_count;
 518     }
 519
 520 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 521 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 522
 523     if((ret=realloc_audio(&s->postin, in_count))<0)
 524         return ret;
 525     if(s->resample_first){
 526         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 527         if((ret=realloc_audio(&s->midbuf, out_count))<0)
 528             return ret;
 529     }else{
 530         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 531         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
 532             return ret;
 533     }
 534     if((ret=realloc_audio(&s->preout, out_count))<0)
 535         return ret;
 536
 537     postin= &s->postin;
 538
 539     midbuf_tmp= s->midbuf;
 540     midbuf= &midbuf_tmp;
 541     preout_tmp= s->preout;
 542     preout= &preout_tmp;
 543
 544     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
 545         postin= in;
 546
 547     if(s->resample_first ? !s->resample : !s->rematrix)
 548         midbuf= postin;
 549
 550     if(s->resample_first ? !s->rematrix : !s->resample)
 551         preout= midbuf;
 552
 553     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 554         if(preout==in){
 555             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 556             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 557             copy(out, in, out_count);
 558             return out_count;
 559         }
 560         else if(preout==postin) preout= midbuf= postin= out;
 561         else if(preout==midbuf) preout= midbuf= out;
 562         else                    preout= out;
 563     }
 564
 565     if(in != postin){
 566         swri_audio_convert(s->in_convert, postin, in, in_count);
 567     }
 568
 569     if(s->resample_first){
 570         if(postin != midbuf)
 571             out_count= resample(s, midbuf, out_count, postin, in_count);
 572         if(midbuf != preout)
 573             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 574     }else{
 575         if(postin != midbuf)
 576             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 577         if(midbuf != preout)
 578             out_count= resample(s, preout, out_count, midbuf, in_count);
 579     }
 580
 581     if(preout != out && out_count){
 582         if(s->dither_method){
 583             int ch;
 584             int dither_count= FFMAX(out_count, 1<<16);
 585             av_assert0(preout != in);
 586
 587             if((ret=realloc_audio(&s->dither, dither_count))<0)
 588                 return ret;
 589             if(ret)
 590                 for(ch=0; ch<s->dither.ch_count; ch++)
 591                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
 592             av_assert0(s->dither.ch_count == preout->ch_count);
 593
 594             if(s->dither_pos + out_count > s->dither.count)
 595                 s->dither_pos = 0;
 596
 597             for(ch=0; ch<preout->ch_count; ch++)
 598                 s->mix_2_1_f(preout->ch[ch], preout->ch[ch], s->dither.ch[ch] + s->dither.bps * s->dither_pos, s->native_one, 0, 0, out_count);
 599
 600             s->dither_pos += out_count;
 601         }
 602 //FIXME packed doesnt need more than 1 chan here!
 603         swri_audio_convert(s->out_convert, out, preout, out_count);
 604     }
 605     return out_count;
 606 }
 607
 608 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 609                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 610     AudioData * in= &s->in;
 611     AudioData *out= &s->out;
 612
 613     if(s->drop_output > 0){
 614         int ret;
 615         AudioData tmp = s->out;
 616         uint8_t *tmp_arg[SWR_CH_MAX];
 617         tmp.count = 0;
 618         tmp.data  = NULL;
 619         if((ret=realloc_audio(&tmp, s->drop_output))<0)
 620             return ret;
 621
 622         reversefill_audiodata(&tmp, tmp_arg);
 623         s->drop_output *= -1; //FIXME find a less hackish solution
 624         ret = swr_convert(s, tmp_arg, -s->drop_output, in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesnt matter
 625         s->drop_output *= -1;
 626         if(ret>0)
 627             s->drop_output -= ret;
 628
 629         av_freep(&tmp.data);
 630         if(s->drop_output || !out_arg)
 631             return 0;
 632         in_count = 0;
 633     }
 634
 635     if(!in_arg){
 636         if(s->in_buffer_count){
 637             if (s->resample && !s->flushed) {
 638                 AudioData *a= &s->in_buffer;
 639                 int i, j, ret;
 640                 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
 641                     return ret;
 642                 av_assert0(a->planar);
 643                 for(i=0; i<a->ch_count; i++){
 644                     for(j=0; j<s->in_buffer_count; j++){
 645                         memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
 646                             a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
 647                     }
 648                 }
 649                 s->in_buffer_count += (s->in_buffer_count+1)/2;
 650                 s->resample_in_constraint = 0;
 651                 s->flushed = 1;
 652             }
 653         }else{
 654             return 0;
 655         }
 656     }else
 657         fill_audiodata(in ,  (void*)in_arg);
 658
 659     fill_audiodata(out, out_arg);
 660
 661     if(s->resample){
 662         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 663         if(ret>0 && !s->drop_output)
 664             s->outpts += ret * (int64_t)s->in_sample_rate;
 665         return ret;
 666     }else{
 667         AudioData tmp= *in;
 668         int ret2=0;
 669         int ret, size;
 670         size = FFMIN(out_count, s->in_buffer_count);
 671         if(size){
 672             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 673             ret= swr_convert_internal(s, out, size, &tmp, size);
 674             if(ret<0)
 675                 return ret;
 676             ret2= ret;
 677             s->in_buffer_count -= ret;
 678             s->in_buffer_index += ret;
 679             buf_set(out, out, ret);
 680             out_count -= ret;
 681             if(!s->in_buffer_count)
 682                 s->in_buffer_index = 0;
 683         }
 684
 685         if(in_count){
 686             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 687
 688             if(in_count > out_count) { //FIXME move after swr_convert_internal
 689                 if(   size > s->in_buffer.count
 690                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 691                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 692                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 693                     s->in_buffer_index=0;
 694                 }else
 695                     if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 696                         return ret;
 697             }
 698
 699             if(out_count){
 700                 size = FFMIN(in_count, out_count);
 701                 ret= swr_convert_internal(s, out, size, in, size);
 702                 if(ret<0)
 703                     return ret;
 704                 buf_set(in, in, ret);
 705                 in_count -= ret;
 706                 ret2 += ret;
 707             }
 708             if(in_count){
 709                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 710                 copy(&tmp, in, in_count);
 711                 s->in_buffer_count += in_count;
 712             }
 713         }
 714         if(ret2>0 && !s->drop_output)
 715             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 716         return ret2;
 717     }
 718 }
 719
 720 int swr_drop_output(struct SwrContext *s, int count){
 721     s->drop_output += count;
 722
 723     if(s->drop_output <= 0)
 724         return 0;
 725
 726     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 727     return swr_convert(s, NULL, s->drop_output, NULL, 0);
 728 }
 729
 730 int swr_inject_silence(struct SwrContext *s, int count){
 731     int ret, i;
 732     AudioData silence = s->out;
 733     uint8_t *tmp_arg[SWR_CH_MAX];
 734
 735     if(count <= 0)
 736         return 0;
 737
 738     silence.count = 0;
 739     silence.data  = NULL;
 740     if((ret=realloc_audio(&silence, count))<0)
 741         return ret;
 742
 743     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
 744         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
 745     } else
 746         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
 747
 748     reversefill_audiodata(&silence, tmp_arg);
 749     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 750     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 751     av_freep(&silence.data);
 752     return ret;
 753 }
 754
 755 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 756     if(pts == INT64_MIN)
 757         return s->outpts;
 758     if(s->min_compensation >= FLT_MAX) {
 759         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 760     } else {
 761         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
 762         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 763
 764         if(fabs(fdelta) > s->min_compensation) {
 765             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
 766                 int ret;
 767                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 768                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 769                 if(ret<0){
 770                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 771                 }
 772             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 773                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 774                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 775                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 776                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 777                 swr_set_compensation(s, comp, duration);
 778             }
 779         }
 780
 781         return s->outpts;
 782     }
 783 }