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

   1 /*
   2  * Copyright (C) 2011-2013 Michael Niedermayer (michaelni@gmx.at)
   3  *
   4  * This file is part of libswresample
   5  *
   6  * libswresample is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * libswresample is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with libswresample; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "libavutil/opt.h"
  22 #include "swresample_internal.h"
  23 #include "audioconvert.h"
  24 #include "libavutil/avassert.h"
  25 #include "libavutil/channel_layout.h"
  26
  27 #include <float.h>
  28
  29 #define  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"                  , "set input channel count"     , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  45 {"in_channel_count"     , "set input channel count"     , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  46 {"och"                  , "set output channel count"    , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  47 {"out_channel_count"    , "set output channel count"    , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  48 {"uch"                  , "set used channel count"      , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  49 {"used_channel_count"   , "set used channel count"      , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  50 {"isr"                  , "set input sample rate"       , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  51 {"in_sample_rate"       , "set input sample rate"       , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  52 {"osr"                  , "set output sample rate"      , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  53 {"out_sample_rate"      , "set output sample rate"      , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  54 {"isf"                  , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
  55 {"in_sample_fmt"        , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
  56 {"osf"                  , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
  57 {"out_sample_fmt"       , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
  58 {"tsf"                  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
  59 {"internal_sample_fmt"  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_SAMPLE_FMT , {.i64=AV_SAMPLE_FMT_NONE}, -1   , AV_SAMPLE_FMT_NB-1, PARAM},
  60 {"icl"                  , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  61 {"in_channel_layout"    , "set input channel layout"    , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  62 {"ocl"                  , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  63 {"out_channel_layout"   , "set output channel layout"   , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  64 {"clev"                 , "set center mix level"        , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  65 {"center_mix_level"     , "set center mix level"        , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  66 {"slev"                 , "set surround mix level"      , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  67 {"surround_mix_level"   , "set surround mix Level"      , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  68 {"lfe_mix_level"        , "set LFE mix level"           , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
  69 {"rmvol"                , "set rematrix volume"         , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
  70 {"rematrix_volume"      , "set rematrix volume"         , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
  71
  72 {"flags"                , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
  73 {"swr_flags"            , "set flags"                   , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
  74 {"res"                  , "force resampling"            , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
  75
  76 {"dither_scale"         , "set dither scale"            , OFFSET(dither.scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
  77
  78 {"dither_method"        , "set dither method"           , OFFSET(dither.method  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
  79 {"rectangular"          , "select rectangular dither"   , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
  80 {"triangular"           , "select triangular dither"    , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
  81 {"triangular_hp"        , "select triangular dither with high pass" , 0          , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  82 {"lipshitz"             , "select lipshitz noise shaping dither" , 0             , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_LIPSHITZ}, INT_MIN, INT_MAX, PARAM, "dither_method"},
  83 {"shibata"              , "select shibata noise shaping dither" , 0              , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  84 {"low_shibata"          , "select low shibata noise shaping dither" , 0          , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_LOW_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  85 {"high_shibata"         , "select high shibata noise shaping dither" , 0         , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_HIGH_SHIBATA }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  86 {"f_weighted"           , "select f-weighted noise shaping dither" , 0           , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_F_WEIGHTED }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  87 {"modified_e_weighted"  , "select modified-e-weighted noise shaping dither" , 0  , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_MODIFIED_E_WEIGHTED }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  88 {"improved_e_weighted"  , "select improved-e-weighted noise shaping dither" , 0  , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_NS_IMPROVED_E_WEIGHTED }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  89
  90 {"filter_size"          , "set swr resampling filter size", OFFSET(filter_size)  , AV_OPT_TYPE_INT  , {.i64=32                    }, 0      , INT_MAX   , PARAM },
  91 {"phase_shift"          , "set swr resampling phase shift", OFFSET(phase_shift)  , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 24        , PARAM },
  92 {"linear_interp"        , "enable linear interpolation" , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
  93 {"cutoff"               , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
  94 {"resampler"            , "set resampling Engine"       , OFFSET(engine)         , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_ENGINE_NB-1, PARAM, "resampler"},
  95 {"swr"                  , "select SW Resampler"         , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SWR        }, INT_MIN, INT_MAX   , PARAM, "resampler"},
  96 {"soxr"                 , "select SoX Resampler"        , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SOXR       }, INT_MIN, INT_MAX   , PARAM, "resampler"},
  97 {"precision"            , "set soxr resampling precision (in bits)"
  98                                                         , OFFSET(precision)      , AV_OPT_TYPE_DOUBLE,{.dbl=20.0                  }, 15.0   , 33.0      , PARAM },
  99 {"cheby"                , "enable soxr Chebyshev passband & higher-precision irrational ratio approximation"
 100                                                         , OFFSET(cheby)          , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
 101 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
 102                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
 103 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
 104                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
 105 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
 106                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
 107 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
 108                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
 109 {"async"                , "simplified 1 parameter audio timestamp matching, 0(disabled), 1(filling and trimming), >1(maximum stretch/squeeze in samples per second)"
 110                                                         , OFFSET(async)          , AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
 111 {"first_pts"            , "Assume the first pts should be this value (in samples)."
 112                                                         , OFFSET(firstpts_in_samples), AV_OPT_TYPE_INT64 ,{.i64=AV_NOPTS_VALUE    }, INT64_MIN,INT64_MAX, PARAM },
 113
 114 { "matrix_encoding"     , "set matrixed stereo encoding" , OFFSET(matrix_encoding), AV_OPT_TYPE_INT   ,{.i64 = AV_MATRIX_ENCODING_NONE}, AV_MATRIX_ENCODING_NONE,     AV_MATRIX_ENCODING_NB-1, PARAM, "matrix_encoding" },
 115     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 116     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 117     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 118
 119 { "filter_type"         , "select swr filter type"      , OFFSET(filter_type)    , AV_OPT_TYPE_INT  , { .i64 = SWR_FILTER_TYPE_KAISER }, SWR_FILTER_TYPE_CUBIC, SWR_FILTER_TYPE_KAISER, PARAM, "filter_type" },
 120     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 121     { "blackman_nuttall", "select Blackman Nuttall Windowed Sinc", 0             , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_BLACKMAN_NUTTALL }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 122     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 123
 124 { "kaiser_beta"         , "set swr Kaiser Window Beta"  , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
 125
 126 {0}
 127 };
 128
 129 static const char* context_to_name(void* ptr) {
 130     return "SWR";
 131 }
 132
 133 static const AVClass av_class = {
 134     .class_name                = "SWResampler",
 135     .item_name                 = context_to_name,
 136     .option                    = options,
 137     .version                   = LIBAVUTIL_VERSION_INT,
 138     .log_level_offset_offset   = OFFSET(log_level_offset),
 139     .parent_log_context_offset = OFFSET(log_ctx),
 140     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
 141 };
 142
 143 unsigned swresample_version(void)
 144 {
 145     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
 146     return LIBSWRESAMPLE_VERSION_INT;
 147 }
 148
 149 const char *swresample_configuration(void)
 150 {
 151     return FFMPEG_CONFIGURATION;
 152 }
 153
 154 const char *swresample_license(void)
 155 {
 156 #define LICENSE_PREFIX "libswresample license: "
 157     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
 158 }
 159
 160 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
 161     if(!s || s->in_convert) // s needs to be allocated but not initialized
 162         return AVERROR(EINVAL);
 163     s->channel_map = channel_map;
 164     return 0;
 165 }
 166
 167 const AVClass *swr_get_class(void)
 168 {
 169     return &av_class;
 170 }
 171
 172 av_cold struct SwrContext *swr_alloc(void){
 173     SwrContext *s= av_mallocz(sizeof(SwrContext));
 174     if(s){
 175         s->av_class= &av_class;
 176         av_opt_set_defaults(s);
 177     }
 178     return s;
 179 }
 180
 181 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
 182                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
 183                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
 184                                       int log_offset, void *log_ctx){
 185     if(!s) s= swr_alloc();
 186     if(!s) return NULL;
 187
 188     s->log_level_offset= log_offset;
 189     s->log_ctx= log_ctx;
 190
 191     av_opt_set_int(s, "ocl", out_ch_layout,   0);
 192     av_opt_set_int(s, "osf", out_sample_fmt,  0);
 193     av_opt_set_int(s, "osr", out_sample_rate, 0);
 194     av_opt_set_int(s, "icl", in_ch_layout,    0);
 195     av_opt_set_int(s, "isf", in_sample_fmt,   0);
 196     av_opt_set_int(s, "isr", in_sample_rate,  0);
 197     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
 198     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
 199     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
 200     av_opt_set_int(s, "uch", 0, 0);
 201     return s;
 202 }
 203
 204 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 205     a->fmt   = fmt;
 206     a->bps   = av_get_bytes_per_sample(fmt);
 207     a->planar= av_sample_fmt_is_planar(fmt);
 208 }
 209
 210 static void free_temp(AudioData *a){
 211     av_free(a->data);
 212     memset(a, 0, sizeof(*a));
 213 }
 214
 215 av_cold void swr_free(SwrContext **ss){
 216     SwrContext *s= *ss;
 217     if(s){
 218         free_temp(&s->postin);
 219         free_temp(&s->midbuf);
 220         free_temp(&s->preout);
 221         free_temp(&s->in_buffer);
 222         free_temp(&s->silence);
 223         free_temp(&s->drop_temp);
 224         free_temp(&s->dither.noise);
 225         free_temp(&s->dither.temp);
 226         swri_audio_convert_free(&s-> in_convert);
 227         swri_audio_convert_free(&s->out_convert);
 228         swri_audio_convert_free(&s->full_convert);
 229         if (s->resampler)
 230             s->resampler->free(&s->resample);
 231         swri_rematrix_free(s);
 232     }
 233
 234     av_freep(ss);
 235 }
 236
 237 av_cold int swr_init(struct SwrContext *s){
 238     int ret;
 239     s->in_buffer_index= 0;
 240     s->in_buffer_count= 0;
 241     s->resample_in_constraint= 0;
 242     free_temp(&s->postin);
 243     free_temp(&s->midbuf);
 244     free_temp(&s->preout);
 245     free_temp(&s->in_buffer);
 246     free_temp(&s->silence);
 247     free_temp(&s->drop_temp);
 248     free_temp(&s->dither.noise);
 249     free_temp(&s->dither.temp);
 250     memset(s->in.ch, 0, sizeof(s->in.ch));
 251     memset(s->out.ch, 0, sizeof(s->out.ch));
 252     swri_audio_convert_free(&s-> in_convert);
 253     swri_audio_convert_free(&s->out_convert);
 254     swri_audio_convert_free(&s->full_convert);
 255     swri_rematrix_free(s);
 256
 257     s->flushed = 0;
 258
 259     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 260         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 261         return AVERROR(EINVAL);
 262     }
 263     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 264         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 265         return AVERROR(EINVAL);
 266     }
 267
 268     switch(s->engine){
 269 #if CONFIG_LIBSOXR
 270         extern struct Resampler const soxr_resampler;
 271         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
 272 #endif
 273         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
 274         default:
 275             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
 276             return AVERROR(EINVAL);
 277     }
 278
 279     if(!s->used_ch_count)
 280         s->used_ch_count= s->in.ch_count;
 281
 282     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 283         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 284         s-> in_ch_layout= 0;
 285     }
 286
 287     if(!s-> in_ch_layout)
 288         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 289     if(!s->out_ch_layout)
 290         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 291
 292     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 293                  s->rematrix_custom;
 294
 295     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 296         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 297             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 298         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
 299                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
 300                  && !s->rematrix
 301                  && s->engine != SWR_ENGINE_SOXR){
 302             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
 303         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 304             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 305         }else{
 306             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 307             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 308         }
 309     }
 310
 311     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 312         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 313         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 314         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 315         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));
 316         return AVERROR(EINVAL);
 317     }
 318
 319     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 320     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 321
 322     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
 323         if (!s->async && s->min_compensation >= FLT_MAX/2)
 324             s->async = 1;
 325         s->firstpts =
 326         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
 327     }
 328
 329     if (s->async) {
 330         if (s->min_compensation >= FLT_MAX/2)
 331             s->min_compensation = 0.001;
 332         if (s->async > 1.0001) {
 333             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
 334         }
 335     }
 336
 337     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 338         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);
 339     }else
 340         s->resampler->free(&s->resample);
 341     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 342         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 343         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 344         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 345         && s->resample){
 346         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 347         return -1;
 348     }
 349
 350 #define RSC 1 //FIXME finetune
 351     if(!s-> in.ch_count)
 352         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 353     if(!s->used_ch_count)
 354         s->used_ch_count= s->in.ch_count;
 355     if(!s->out.ch_count)
 356         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 357
 358     if(!s-> in.ch_count){
 359         av_assert0(!s->in_ch_layout);
 360         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 361         return -1;
 362     }
 363
 364     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 365         char l1[1024], l2[1024];
 366         av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
 367         av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
 368         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
 369                "but there is not enough information to do it\n", l1, l2);
 370         return -1;
 371     }
 372
 373 av_assert0(s->used_ch_count);
 374 av_assert0(s->out.ch_count);
 375     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 376
 377     s->in_buffer= s->in;
 378     s->silence  = s->in;
 379     s->drop_temp= s->out;
 380
 381     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
 382         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 383                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 384         return 0;
 385     }
 386
 387     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 388                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 389     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 390                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 391
 392     if (!s->in_convert || !s->out_convert)
 393         return AVERROR(ENOMEM);
 394
 395     s->postin= s->in;
 396     s->preout= s->out;
 397     s->midbuf= s->in;
 398
 399     if(s->channel_map){
 400         s->postin.ch_count=
 401         s->midbuf.ch_count= s->used_ch_count;
 402         if(s->resample)
 403             s->in_buffer.ch_count= s->used_ch_count;
 404     }
 405     if(!s->resample_first){
 406         s->midbuf.ch_count= s->out.ch_count;
 407         if(s->resample)
 408             s->in_buffer.ch_count = s->out.ch_count;
 409     }
 410
 411     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 412     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 413     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 414
 415     if(s->resample){
 416         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 417     }
 418
 419     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
 420         return ret;
 421
 422     if(s->rematrix || s->dither.method)
 423         return swri_rematrix_init(s);
 424
 425     return 0;
 426 }
 427
 428 int swri_realloc_audio(AudioData *a, int count){
 429     int i, countb;
 430     AudioData old;
 431
 432     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 433         return AVERROR(EINVAL);
 434
 435     if(a->count >= count)
 436         return 0;
 437
 438     count*=2;
 439
 440     countb= FFALIGN(count*a->bps, ALIGN);
 441     old= *a;
 442
 443     av_assert0(a->bps);
 444     av_assert0(a->ch_count);
 445
 446     a->data= av_mallocz(countb*a->ch_count);
 447     if(!a->data)
 448         return AVERROR(ENOMEM);
 449     for(i=0; i<a->ch_count; i++){
 450         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 451         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 452     }
 453     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 454     av_free(old.data);
 455     a->count= count;
 456
 457     return 1;
 458 }
 459
 460 static void copy(AudioData *out, AudioData *in,
 461                  int count){
 462     av_assert0(out->planar == in->planar);
 463     av_assert0(out->bps == in->bps);
 464     av_assert0(out->ch_count == in->ch_count);
 465     if(out->planar){
 466         int ch;
 467         for(ch=0; ch<out->ch_count; ch++)
 468             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 469     }else
 470         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 471 }
 472
 473 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 474     int i;
 475     if(!in_arg){
 476         memset(out->ch, 0, sizeof(out->ch));
 477     }else if(out->planar){
 478         for(i=0; i<out->ch_count; i++)
 479             out->ch[i]= in_arg[i];
 480     }else{
 481         for(i=0; i<out->ch_count; i++)
 482             out->ch[i]= in_arg[0] + i*out->bps;
 483     }
 484 }
 485
 486 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 487     int i;
 488     if(out->planar){
 489         for(i=0; i<out->ch_count; i++)
 490             in_arg[i]= out->ch[i];
 491     }else{
 492         in_arg[0]= out->ch[0];
 493     }
 494 }
 495
 496 /**
 497  *
 498  * out may be equal in.
 499  */
 500 static void buf_set(AudioData *out, AudioData *in, int count){
 501     int ch;
 502     if(in->planar){
 503         for(ch=0; ch<out->ch_count; ch++)
 504             out->ch[ch]= in->ch[ch] + count*out->bps;
 505     }else{
 506         for(ch=out->ch_count-1; ch>=0; ch--)
 507             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 508     }
 509 }
 510
 511 /**
 512  *
 513  * @return number of samples output per channel
 514  */
 515 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 516                              const AudioData * in_param, int in_count){
 517     AudioData in, out, tmp;
 518     int ret_sum=0;
 519     int border=0;
 520
 521     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 522     av_assert1(s->in_buffer.planar   == in_param->planar);
 523     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 524
 525     tmp=out=*out_param;
 526     in =  *in_param;
 527
 528     do{
 529         int ret, size, consumed;
 530         if(!s->resample_in_constraint && s->in_buffer_count){
 531             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 532             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 533             out_count -= ret;
 534             ret_sum += ret;
 535             buf_set(&out, &out, ret);
 536             s->in_buffer_count -= consumed;
 537             s->in_buffer_index += consumed;
 538
 539             if(!in_count)
 540                 break;
 541             if(s->in_buffer_count <= border){
 542                 buf_set(&in, &in, -s->in_buffer_count);
 543                 in_count += s->in_buffer_count;
 544                 s->in_buffer_count=0;
 545                 s->in_buffer_index=0;
 546                 border = 0;
 547             }
 548         }
 549
 550         if((s->flushed || in_count) && !s->in_buffer_count){
 551             s->in_buffer_index=0;
 552             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 553             out_count -= ret;
 554             ret_sum += ret;
 555             buf_set(&out, &out, ret);
 556             in_count -= consumed;
 557             buf_set(&in, &in, consumed);
 558         }
 559
 560         //TODO is this check sane considering the advanced copy avoidance below
 561         size= s->in_buffer_index + s->in_buffer_count + in_count;
 562         if(   size > s->in_buffer.count
 563            && s->in_buffer_count + in_count <= s->in_buffer_index){
 564             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 565             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 566             s->in_buffer_index=0;
 567         }else
 568             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 569                 return ret;
 570
 571         if(in_count){
 572             int count= in_count;
 573             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 574
 575             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 576             copy(&tmp, &in, /*in_*/count);
 577             s->in_buffer_count += count;
 578             in_count -= count;
 579             border += count;
 580             buf_set(&in, &in, count);
 581             s->resample_in_constraint= 0;
 582             if(s->in_buffer_count != count || in_count)
 583                 continue;
 584         }
 585         break;
 586     }while(1);
 587
 588     s->resample_in_constraint= !!out_count;
 589
 590     return ret_sum;
 591 }
 592
 593 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 594                                                       AudioData *in , int  in_count){
 595     AudioData *postin, *midbuf, *preout;
 596     int ret/*, in_max*/;
 597     AudioData preout_tmp, midbuf_tmp;
 598
 599     if(s->full_convert){
 600         av_assert0(!s->resample);
 601         swri_audio_convert(s->full_convert, out, in, in_count);
 602         return out_count;
 603     }
 604
 605 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 606 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 607
 608     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
 609         return ret;
 610     if(s->resample_first){
 611         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 612         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
 613             return ret;
 614     }else{
 615         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 616         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
 617             return ret;
 618     }
 619     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
 620         return ret;
 621
 622     postin= &s->postin;
 623
 624     midbuf_tmp= s->midbuf;
 625     midbuf= &midbuf_tmp;
 626     preout_tmp= s->preout;
 627     preout= &preout_tmp;
 628
 629     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 630         postin= in;
 631
 632     if(s->resample_first ? !s->resample : !s->rematrix)
 633         midbuf= postin;
 634
 635     if(s->resample_first ? !s->rematrix : !s->resample)
 636         preout= midbuf;
 637
 638     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 639         if(preout==in){
 640             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 641             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 642             copy(out, in, out_count);
 643             return out_count;
 644         }
 645         else if(preout==postin) preout= midbuf= postin= out;
 646         else if(preout==midbuf) preout= midbuf= out;
 647         else                    preout= out;
 648     }
 649
 650     if(in != postin){
 651         swri_audio_convert(s->in_convert, postin, in, in_count);
 652     }
 653
 654     if(s->resample_first){
 655         if(postin != midbuf)
 656             out_count= resample(s, midbuf, out_count, postin, in_count);
 657         if(midbuf != preout)
 658             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 659     }else{
 660         if(postin != midbuf)
 661             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 662         if(midbuf != preout)
 663             out_count= resample(s, preout, out_count, midbuf, in_count);
 664     }
 665
 666     if(preout != out && out_count){
 667         AudioData *conv_src = preout;
 668         if(s->dither.method){
 669             int ch;
 670             int dither_count= FFMAX(out_count, 1<<16);
 671
 672             if (preout == in) {
 673                 conv_src = &s->dither.temp;
 674                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
 675                     return ret;
 676             }
 677
 678             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
 679                 return ret;
 680             if(ret)
 681                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
 682                     swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
 683             av_assert0(s->dither.noise.ch_count == preout->ch_count);
 684
 685             if(s->dither.noise_pos + out_count > s->dither.noise.count)
 686                 s->dither.noise_pos = 0;
 687
 688             if (s->dither.method < SWR_DITHER_NS){
 689                 if (s->mix_2_1_simd) {
 690                     int len1= out_count&~15;
 691                     int off = len1 * preout->bps;
 692
 693                     if(len1)
 694                         for(ch=0; ch<preout->ch_count; ch++)
 695                             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_one, 0, 0, len1);
 696                     if(out_count != len1)
 697                         for(ch=0; ch<preout->ch_count; ch++)
 698                             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);
 699                 } else {
 700                     for(ch=0; ch<preout->ch_count; ch++)
 701                         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);
 702                 }
 703             } else {
 704                 switch(s->int_sample_fmt) {
 705                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
 706                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
 707                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
 708                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
 709                 }
 710             }
 711             s->dither.noise_pos += out_count;
 712         }
 713 //FIXME packed doesnt need more than 1 chan here!
 714         swri_audio_convert(s->out_convert, out, conv_src, out_count);
 715     }
 716     return out_count;
 717 }
 718
 719 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 720                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 721     AudioData * in= &s->in;
 722     AudioData *out= &s->out;
 723
 724     while(s->drop_output > 0){
 725         int ret;
 726         uint8_t *tmp_arg[SWR_CH_MAX];
 727 #define MAX_DROP_STEP 16384
 728         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
 729             return ret;
 730
 731         reversefill_audiodata(&s->drop_temp, tmp_arg);
 732         s->drop_output *= -1; //FIXME find a less hackish solution
 733         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 doesnt matter
 734         s->drop_output *= -1;
 735         in_count = 0;
 736         if(ret>0) {
 737             s->drop_output -= ret;
 738             continue;
 739         }
 740
 741         if(s->drop_output || !out_arg)
 742             return 0;
 743     }
 744
 745     if(!in_arg){
 746         if(s->resample){
 747             if (!s->flushed)
 748                 s->resampler->flush(s);
 749             s->resample_in_constraint = 0;
 750             s->flushed = 1;
 751         }else if(!s->in_buffer_count){
 752             return 0;
 753         }
 754     }else
 755         fill_audiodata(in ,  (void*)in_arg);
 756
 757     fill_audiodata(out, out_arg);
 758
 759     if(s->resample){
 760         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 761         if(ret>0 && !s->drop_output)
 762             s->outpts += ret * (int64_t)s->in_sample_rate;
 763         return ret;
 764     }else{
 765         AudioData tmp= *in;
 766         int ret2=0;
 767         int ret, size;
 768         size = FFMIN(out_count, s->in_buffer_count);
 769         if(size){
 770             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 771             ret= swr_convert_internal(s, out, size, &tmp, size);
 772             if(ret<0)
 773                 return ret;
 774             ret2= ret;
 775             s->in_buffer_count -= ret;
 776             s->in_buffer_index += ret;
 777             buf_set(out, out, ret);
 778             out_count -= ret;
 779             if(!s->in_buffer_count)
 780                 s->in_buffer_index = 0;
 781         }
 782
 783         if(in_count){
 784             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 785
 786             if(in_count > out_count) { //FIXME move after swr_convert_internal
 787                 if(   size > s->in_buffer.count
 788                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 789                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 790                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 791                     s->in_buffer_index=0;
 792                 }else
 793                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 794                         return ret;
 795             }
 796
 797             if(out_count){
 798                 size = FFMIN(in_count, out_count);
 799                 ret= swr_convert_internal(s, out, size, in, size);
 800                 if(ret<0)
 801                     return ret;
 802                 buf_set(in, in, ret);
 803                 in_count -= ret;
 804                 ret2 += ret;
 805             }
 806             if(in_count){
 807                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 808                 copy(&tmp, in, in_count);
 809                 s->in_buffer_count += in_count;
 810             }
 811         }
 812         if(ret2>0 && !s->drop_output)
 813             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 814         return ret2;
 815     }
 816 }
 817
 818 int swr_drop_output(struct SwrContext *s, int count){
 819     s->drop_output += count;
 820
 821     if(s->drop_output <= 0)
 822         return 0;
 823
 824     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 825     return swr_convert(s, NULL, s->drop_output, NULL, 0);
 826 }
 827
 828 int swr_inject_silence(struct SwrContext *s, int count){
 829     int ret, i;
 830     uint8_t *tmp_arg[SWR_CH_MAX];
 831
 832     if(count <= 0)
 833         return 0;
 834
 835 #define MAX_SILENCE_STEP 16384
 836     while (count > MAX_SILENCE_STEP) {
 837         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
 838             return ret;
 839         count -= MAX_SILENCE_STEP;
 840     }
 841
 842     if((ret=swri_realloc_audio(&s->silence, count))<0)
 843         return ret;
 844
 845     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
 846         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
 847     } else
 848         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
 849
 850     reversefill_audiodata(&s->silence, tmp_arg);
 851     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 852     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 853     return ret;
 854 }
 855
 856 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
 857     if (s->resampler && s->resample){
 858         return s->resampler->get_delay(s, base);
 859     }else{
 860         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
 861     }
 862 }
 863
 864 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
 865     int ret;
 866
 867     if (!s || compensation_distance < 0)
 868         return AVERROR(EINVAL);
 869     if (!compensation_distance && sample_delta)
 870         return AVERROR(EINVAL);
 871     if (!s->resample) {
 872         s->flags |= SWR_FLAG_RESAMPLE;
 873         ret = swr_init(s);
 874         if (ret < 0)
 875             return ret;
 876     }
 877     if (!s->resampler->set_compensation){
 878         return AVERROR(EINVAL);
 879     }else{
 880         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
 881     }
 882 }
 883
 884 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 885     if(pts == INT64_MIN)
 886         return s->outpts;
 887     if(s->min_compensation >= FLT_MAX) {
 888         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 889     } else {
 890         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;
 891         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 892
 893         if(fabs(fdelta) > s->min_compensation) {
 894             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
 895                 int ret;
 896                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 897                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 898                 if(ret<0){
 899                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 900                 }
 901             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 902                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 903                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 904                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 905                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 906                 swr_set_compensation(s, comp, duration);
 907             }
 908         }
 909
 910         return s->outpts;
 911     }
 912 }