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     if(av_get_channel_layout_nb_channels(s-> in_ch_layout) > SWR_CH_MAX) {
 269         av_log(s, AV_LOG_WARNING, "Input channel layout 0x%"PRIx64" is invalid or unsupported.\n", s-> in_ch_layout);
 270         s->in_ch_layout = 0;
 271     }
 272
 273     if(av_get_channel_layout_nb_channels(s->out_ch_layout) > SWR_CH_MAX) {
 274         av_log(s, AV_LOG_WARNING, "Output channel layout 0x%"PRIx64" is invalid or unsupported.\n", s->out_ch_layout);
 275         s->out_ch_layout = 0;
 276     }
 277
 278     switch(s->engine){
 279 #if CONFIG_LIBSOXR
 280         extern struct Resampler const soxr_resampler;
 281         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
 282 #endif
 283         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
 284         default:
 285             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
 286             return AVERROR(EINVAL);
 287     }
 288
 289     if(!s->used_ch_count)
 290         s->used_ch_count= s->in.ch_count;
 291
 292     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 293         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 294         s-> in_ch_layout= 0;
 295     }
 296
 297     if(!s-> in_ch_layout)
 298         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 299     if(!s->out_ch_layout)
 300         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 301
 302     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 303                  s->rematrix_custom;
 304
 305     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 306         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 307             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 308         }else if(   av_get_planar_sample_fmt(s-> in_sample_fmt) == AV_SAMPLE_FMT_S32P
 309                  && av_get_planar_sample_fmt(s->out_sample_fmt) == AV_SAMPLE_FMT_S32P
 310                  && !s->rematrix
 311                  && s->engine != SWR_ENGINE_SOXR){
 312             s->int_sample_fmt= AV_SAMPLE_FMT_S32P;
 313         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 314             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 315         }else{
 316             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 317             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 318         }
 319     }
 320
 321     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 322         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 323         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 324         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 325         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));
 326         return AVERROR(EINVAL);
 327     }
 328
 329     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 330     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 331
 332     if (s->firstpts_in_samples != AV_NOPTS_VALUE) {
 333         if (!s->async && s->min_compensation >= FLT_MAX/2)
 334             s->async = 1;
 335         s->firstpts =
 336         s->outpts   = s->firstpts_in_samples * s->out_sample_rate;
 337     }
 338
 339     if (s->async) {
 340         if (s->min_compensation >= FLT_MAX/2)
 341             s->min_compensation = 0.001;
 342         if (s->async > 1.0001) {
 343             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
 344         }
 345     }
 346
 347     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 348         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);
 349     }else
 350         s->resampler->free(&s->resample);
 351     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 352         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 353         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 354         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 355         && s->resample){
 356         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 357         return -1;
 358     }
 359
 360 #define RSC 1 //FIXME finetune
 361     if(!s-> in.ch_count)
 362         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 363     if(!s->used_ch_count)
 364         s->used_ch_count= s->in.ch_count;
 365     if(!s->out.ch_count)
 366         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 367
 368     if(!s-> in.ch_count){
 369         av_assert0(!s->in_ch_layout);
 370         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 371         return -1;
 372     }
 373
 374     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 375         char l1[1024], l2[1024];
 376         av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
 377         av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
 378         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
 379                "but there is not enough information to do it\n", l1, l2);
 380         return -1;
 381     }
 382
 383 av_assert0(s->used_ch_count);
 384 av_assert0(s->out.ch_count);
 385     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 386
 387     s->in_buffer= s->in;
 388     s->silence  = s->in;
 389     s->drop_temp= s->out;
 390
 391     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither.method){
 392         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 393                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 394         return 0;
 395     }
 396
 397     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 398                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 399     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 400                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 401
 402     if (!s->in_convert || !s->out_convert)
 403         return AVERROR(ENOMEM);
 404
 405     s->postin= s->in;
 406     s->preout= s->out;
 407     s->midbuf= s->in;
 408
 409     if(s->channel_map){
 410         s->postin.ch_count=
 411         s->midbuf.ch_count= s->used_ch_count;
 412         if(s->resample)
 413             s->in_buffer.ch_count= s->used_ch_count;
 414     }
 415     if(!s->resample_first){
 416         s->midbuf.ch_count= s->out.ch_count;
 417         if(s->resample)
 418             s->in_buffer.ch_count = s->out.ch_count;
 419     }
 420
 421     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 422     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 423     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 424
 425     if(s->resample){
 426         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 427     }
 428
 429     if ((ret = swri_dither_init(s, s->out_sample_fmt, s->int_sample_fmt)) < 0)
 430         return ret;
 431
 432     if(s->rematrix || s->dither.method)
 433         return swri_rematrix_init(s);
 434
 435     return 0;
 436 }
 437
 438 int swri_realloc_audio(AudioData *a, int count){
 439     int i, countb;
 440     AudioData old;
 441
 442     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 443         return AVERROR(EINVAL);
 444
 445     if(a->count >= count)
 446         return 0;
 447
 448     count*=2;
 449
 450     countb= FFALIGN(count*a->bps, ALIGN);
 451     old= *a;
 452
 453     av_assert0(a->bps);
 454     av_assert0(a->ch_count);
 455
 456     a->data= av_mallocz(countb*a->ch_count);
 457     if(!a->data)
 458         return AVERROR(ENOMEM);
 459     for(i=0; i<a->ch_count; i++){
 460         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 461         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 462     }
 463     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 464     av_free(old.data);
 465     a->count= count;
 466
 467     return 1;
 468 }
 469
 470 static void copy(AudioData *out, AudioData *in,
 471                  int count){
 472     av_assert0(out->planar == in->planar);
 473     av_assert0(out->bps == in->bps);
 474     av_assert0(out->ch_count == in->ch_count);
 475     if(out->planar){
 476         int ch;
 477         for(ch=0; ch<out->ch_count; ch++)
 478             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 479     }else
 480         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 481 }
 482
 483 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 484     int i;
 485     if(!in_arg){
 486         memset(out->ch, 0, sizeof(out->ch));
 487     }else if(out->planar){
 488         for(i=0; i<out->ch_count; i++)
 489             out->ch[i]= in_arg[i];
 490     }else{
 491         for(i=0; i<out->ch_count; i++)
 492             out->ch[i]= in_arg[0] + i*out->bps;
 493     }
 494 }
 495
 496 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 497     int i;
 498     if(out->planar){
 499         for(i=0; i<out->ch_count; i++)
 500             in_arg[i]= out->ch[i];
 501     }else{
 502         in_arg[0]= out->ch[0];
 503     }
 504 }
 505
 506 /**
 507  *
 508  * out may be equal in.
 509  */
 510 static void buf_set(AudioData *out, AudioData *in, int count){
 511     int ch;
 512     if(in->planar){
 513         for(ch=0; ch<out->ch_count; ch++)
 514             out->ch[ch]= in->ch[ch] + count*out->bps;
 515     }else{
 516         for(ch=out->ch_count-1; ch>=0; ch--)
 517             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 518     }
 519 }
 520
 521 /**
 522  *
 523  * @return number of samples output per channel
 524  */
 525 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 526                              const AudioData * in_param, int in_count){
 527     AudioData in, out, tmp;
 528     int ret_sum=0;
 529     int border=0;
 530
 531     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 532     av_assert1(s->in_buffer.planar   == in_param->planar);
 533     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 534
 535     tmp=out=*out_param;
 536     in =  *in_param;
 537
 538     do{
 539         int ret, size, consumed;
 540         if(!s->resample_in_constraint && s->in_buffer_count){
 541             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 542             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 543             out_count -= ret;
 544             ret_sum += ret;
 545             buf_set(&out, &out, ret);
 546             s->in_buffer_count -= consumed;
 547             s->in_buffer_index += consumed;
 548
 549             if(!in_count)
 550                 break;
 551             if(s->in_buffer_count <= border){
 552                 buf_set(&in, &in, -s->in_buffer_count);
 553                 in_count += s->in_buffer_count;
 554                 s->in_buffer_count=0;
 555                 s->in_buffer_index=0;
 556                 border = 0;
 557             }
 558         }
 559
 560         if((s->flushed || in_count) && !s->in_buffer_count){
 561             s->in_buffer_index=0;
 562             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 563             out_count -= ret;
 564             ret_sum += ret;
 565             buf_set(&out, &out, ret);
 566             in_count -= consumed;
 567             buf_set(&in, &in, consumed);
 568         }
 569
 570         //TODO is this check sane considering the advanced copy avoidance below
 571         size= s->in_buffer_index + s->in_buffer_count + in_count;
 572         if(   size > s->in_buffer.count
 573            && s->in_buffer_count + in_count <= s->in_buffer_index){
 574             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 575             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 576             s->in_buffer_index=0;
 577         }else
 578             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 579                 return ret;
 580
 581         if(in_count){
 582             int count= in_count;
 583             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 584
 585             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 586             copy(&tmp, &in, /*in_*/count);
 587             s->in_buffer_count += count;
 588             in_count -= count;
 589             border += count;
 590             buf_set(&in, &in, count);
 591             s->resample_in_constraint= 0;
 592             if(s->in_buffer_count != count || in_count)
 593                 continue;
 594         }
 595         break;
 596     }while(1);
 597
 598     s->resample_in_constraint= !!out_count;
 599
 600     return ret_sum;
 601 }
 602
 603 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 604                                                       AudioData *in , int  in_count){
 605     AudioData *postin, *midbuf, *preout;
 606     int ret/*, in_max*/;
 607     AudioData preout_tmp, midbuf_tmp;
 608
 609     if(s->full_convert){
 610         av_assert0(!s->resample);
 611         swri_audio_convert(s->full_convert, out, in, in_count);
 612         return out_count;
 613     }
 614
 615 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 616 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 617
 618     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
 619         return ret;
 620     if(s->resample_first){
 621         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 622         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
 623             return ret;
 624     }else{
 625         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 626         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
 627             return ret;
 628     }
 629     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
 630         return ret;
 631
 632     postin= &s->postin;
 633
 634     midbuf_tmp= s->midbuf;
 635     midbuf= &midbuf_tmp;
 636     preout_tmp= s->preout;
 637     preout= &preout_tmp;
 638
 639     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 640         postin= in;
 641
 642     if(s->resample_first ? !s->resample : !s->rematrix)
 643         midbuf= postin;
 644
 645     if(s->resample_first ? !s->rematrix : !s->resample)
 646         preout= midbuf;
 647
 648     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 649         if(preout==in){
 650             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 651             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 652             copy(out, in, out_count);
 653             return out_count;
 654         }
 655         else if(preout==postin) preout= midbuf= postin= out;
 656         else if(preout==midbuf) preout= midbuf= out;
 657         else                    preout= out;
 658     }
 659
 660     if(in != postin){
 661         swri_audio_convert(s->in_convert, postin, in, in_count);
 662     }
 663
 664     if(s->resample_first){
 665         if(postin != midbuf)
 666             out_count= resample(s, midbuf, out_count, postin, in_count);
 667         if(midbuf != preout)
 668             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 669     }else{
 670         if(postin != midbuf)
 671             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 672         if(midbuf != preout)
 673             out_count= resample(s, preout, out_count, midbuf, in_count);
 674     }
 675
 676     if(preout != out && out_count){
 677         AudioData *conv_src = preout;
 678         if(s->dither.method){
 679             int ch;
 680             int dither_count= FFMAX(out_count, 1<<16);
 681
 682             if (preout == in) {
 683                 conv_src = &s->dither.temp;
 684                 if((ret=swri_realloc_audio(&s->dither.temp, dither_count))<0)
 685                     return ret;
 686             }
 687
 688             if((ret=swri_realloc_audio(&s->dither.noise, dither_count))<0)
 689                 return ret;
 690             if(ret)
 691                 for(ch=0; ch<s->dither.noise.ch_count; ch++)
 692                     swri_get_dither(s, s->dither.noise.ch[ch], s->dither.noise.count, 12345678913579<<ch, s->dither.noise.fmt);
 693             av_assert0(s->dither.noise.ch_count == preout->ch_count);
 694
 695             if(s->dither.noise_pos + out_count > s->dither.noise.count)
 696                 s->dither.noise_pos = 0;
 697
 698             if (s->dither.method < SWR_DITHER_NS){
 699                 if (s->mix_2_1_simd) {
 700                     int len1= out_count&~15;
 701                     int off = len1 * preout->bps;
 702
 703                     if(len1)
 704                         for(ch=0; ch<preout->ch_count; ch++)
 705                             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);
 706                     if(out_count != len1)
 707                         for(ch=0; ch<preout->ch_count; ch++)
 708                             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);
 709                 } else {
 710                     for(ch=0; ch<preout->ch_count; ch++)
 711                         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);
 712                 }
 713             } else {
 714                 switch(s->int_sample_fmt) {
 715                 case AV_SAMPLE_FMT_S16P :swri_noise_shaping_int16(s, conv_src, preout, &s->dither.noise, out_count); break;
 716                 case AV_SAMPLE_FMT_S32P :swri_noise_shaping_int32(s, conv_src, preout, &s->dither.noise, out_count); break;
 717                 case AV_SAMPLE_FMT_FLTP :swri_noise_shaping_float(s, conv_src, preout, &s->dither.noise, out_count); break;
 718                 case AV_SAMPLE_FMT_DBLP :swri_noise_shaping_double(s,conv_src, preout, &s->dither.noise, out_count); break;
 719                 }
 720             }
 721             s->dither.noise_pos += out_count;
 722         }
 723 //FIXME packed doesnt need more than 1 chan here!
 724         swri_audio_convert(s->out_convert, out, conv_src, out_count);
 725     }
 726     return out_count;
 727 }
 728
 729 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 730                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 731     AudioData * in= &s->in;
 732     AudioData *out= &s->out;
 733
 734     while(s->drop_output > 0){
 735         int ret;
 736         uint8_t *tmp_arg[SWR_CH_MAX];
 737 #define MAX_DROP_STEP 16384
 738         if((ret=swri_realloc_audio(&s->drop_temp, FFMIN(s->drop_output, MAX_DROP_STEP)))<0)
 739             return ret;
 740
 741         reversefill_audiodata(&s->drop_temp, tmp_arg);
 742         s->drop_output *= -1; //FIXME find a less hackish solution
 743         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
 744         s->drop_output *= -1;
 745         in_count = 0;
 746         if(ret>0) {
 747             s->drop_output -= ret;
 748             continue;
 749         }
 750
 751         if(s->drop_output || !out_arg)
 752             return 0;
 753     }
 754
 755     if(!in_arg){
 756         if(s->resample){
 757             if (!s->flushed)
 758                 s->resampler->flush(s);
 759             s->resample_in_constraint = 0;
 760             s->flushed = 1;
 761         }else if(!s->in_buffer_count){
 762             return 0;
 763         }
 764     }else
 765         fill_audiodata(in ,  (void*)in_arg);
 766
 767     fill_audiodata(out, out_arg);
 768
 769     if(s->resample){
 770         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 771         if(ret>0 && !s->drop_output)
 772             s->outpts += ret * (int64_t)s->in_sample_rate;
 773         return ret;
 774     }else{
 775         AudioData tmp= *in;
 776         int ret2=0;
 777         int ret, size;
 778         size = FFMIN(out_count, s->in_buffer_count);
 779         if(size){
 780             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 781             ret= swr_convert_internal(s, out, size, &tmp, size);
 782             if(ret<0)
 783                 return ret;
 784             ret2= ret;
 785             s->in_buffer_count -= ret;
 786             s->in_buffer_index += ret;
 787             buf_set(out, out, ret);
 788             out_count -= ret;
 789             if(!s->in_buffer_count)
 790                 s->in_buffer_index = 0;
 791         }
 792
 793         if(in_count){
 794             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 795
 796             if(in_count > out_count) { //FIXME move after swr_convert_internal
 797                 if(   size > s->in_buffer.count
 798                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 799                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 800                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 801                     s->in_buffer_index=0;
 802                 }else
 803                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 804                         return ret;
 805             }
 806
 807             if(out_count){
 808                 size = FFMIN(in_count, out_count);
 809                 ret= swr_convert_internal(s, out, size, in, size);
 810                 if(ret<0)
 811                     return ret;
 812                 buf_set(in, in, ret);
 813                 in_count -= ret;
 814                 ret2 += ret;
 815             }
 816             if(in_count){
 817                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 818                 copy(&tmp, in, in_count);
 819                 s->in_buffer_count += in_count;
 820             }
 821         }
 822         if(ret2>0 && !s->drop_output)
 823             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 824         return ret2;
 825     }
 826 }
 827
 828 int swr_drop_output(struct SwrContext *s, int count){
 829     s->drop_output += count;
 830
 831     if(s->drop_output <= 0)
 832         return 0;
 833
 834     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 835     return swr_convert(s, NULL, s->drop_output, NULL, 0);
 836 }
 837
 838 int swr_inject_silence(struct SwrContext *s, int count){
 839     int ret, i;
 840     uint8_t *tmp_arg[SWR_CH_MAX];
 841
 842     if(count <= 0)
 843         return 0;
 844
 845 #define MAX_SILENCE_STEP 16384
 846     while (count > MAX_SILENCE_STEP) {
 847         if ((ret = swr_inject_silence(s, MAX_SILENCE_STEP)) < 0)
 848             return ret;
 849         count -= MAX_SILENCE_STEP;
 850     }
 851
 852     if((ret=swri_realloc_audio(&s->silence, count))<0)
 853         return ret;
 854
 855     if(s->silence.planar) for(i=0; i<s->silence.ch_count; i++) {
 856         memset(s->silence.ch[i], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps);
 857     } else
 858         memset(s->silence.ch[0], s->silence.bps==1 ? 0x80 : 0, count*s->silence.bps*s->silence.ch_count);
 859
 860     reversefill_audiodata(&s->silence, tmp_arg);
 861     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 862     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 863     return ret;
 864 }
 865
 866 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
 867     if (s->resampler && s->resample){
 868         return s->resampler->get_delay(s, base);
 869     }else{
 870         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
 871     }
 872 }
 873
 874 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
 875     int ret;
 876
 877     if (!s || compensation_distance < 0)
 878         return AVERROR(EINVAL);
 879     if (!compensation_distance && sample_delta)
 880         return AVERROR(EINVAL);
 881     if (!s->resample) {
 882         s->flags |= SWR_FLAG_RESAMPLE;
 883         ret = swr_init(s);
 884         if (ret < 0)
 885             return ret;
 886     }
 887     if (!s->resampler->set_compensation){
 888         return AVERROR(EINVAL);
 889     }else{
 890         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
 891     }
 892 }
 893
 894 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 895     if(pts == INT64_MIN)
 896         return s->outpts;
 897     if(s->min_compensation >= FLT_MAX) {
 898         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 899     } else {
 900         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;
 901         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 902
 903         if(fabs(fdelta) > s->min_compensation) {
 904             if(s->outpts == s->firstpts || fabs(fdelta) > s->min_hard_compensation){
 905                 int ret;
 906                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 907                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 908                 if(ret<0){
 909                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 910                 }
 911             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 912                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 913                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 914                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 915                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 916                 swr_set_compensation(s, comp, duration);
 917             }
 918         }
 919
 920         return s->outpts;
 921     }
 922 }