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

   1 /*
   2  * Copyright (C) 2011-2012 Michael Niedermayer (michaelni@gmx.at)
   3  *
   4  * This file is part of libswresample
   5  *
   6  * libswresample is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * libswresample is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with libswresample; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "libavutil/opt.h"
  22 #include "swresample_internal.h"
  23 #include "audioconvert.h"
  24 #include "libavutil/avassert.h"
  25 #include "libavutil/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
  83 {"filter_size"          , "set swr resampling filter size", OFFSET(filter_size)  , AV_OPT_TYPE_INT  , {.i64=16                    }, 0      , INT_MAX   , PARAM },
  84 {"phase_shift"          , "set swr resampling phase shift", OFFSET(phase_shift)  , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 30        , PARAM },
  85 {"linear_interp"        , "enable linear interpolation" , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
  86 {"cutoff"               , "set cutoff frequency ratio"  , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.                    }, 0      , 1         , PARAM },
  87 {"resampler"            , "set resampling Engine"       , OFFSET(engine)         , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_ENGINE_NB-1, PARAM, "resampler"},
  88 {"swr"                  , "select SW Resampler"         , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SWR        }, INT_MIN, INT_MAX   , PARAM, "resampler"},
  89 {"soxr"                 , "select SoX Resampler"        , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_ENGINE_SOXR       }, INT_MIN, INT_MAX   , PARAM, "resampler"},
  90 {"precision"            , "set soxr resampling precision (in bits)"
  91                                                         , OFFSET(precision)      , AV_OPT_TYPE_DOUBLE,{.dbl=20.0                  }, 15.0   , 33.0      , PARAM },
  92 {"cheby"                , "enable soxr Chebyshev passband & higher-precision irrational ratio approximation"
  93                                                         , OFFSET(cheby)          , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
  94 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
  95                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
  96 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
  97                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
  98 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
  99                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
 100 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
 101                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
 102 {"async"                , "simplified 1 parameter audio timestamp matching, 0(disabled), 1(filling and trimming), >1(maximum stretch/squeeze in samples per second)"
 103                                                         , OFFSET(async)          , AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
 104
 105 { "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" },
 106     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 107     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 108     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 109
 110 { "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" },
 111     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 112     { "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" },
 113     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 114
 115 { "kaiser_beta"         , "set swr Kaiser Window Beta"  , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
 116
 117 {0}
 118 };
 119
 120 static const char* context_to_name(void* ptr) {
 121     return "SWR";
 122 }
 123
 124 static const AVClass av_class = {
 125     .class_name                = "SWResampler",
 126     .item_name                 = context_to_name,
 127     .option                    = options,
 128     .version                   = LIBAVUTIL_VERSION_INT,
 129     .log_level_offset_offset   = OFFSET(log_level_offset),
 130     .parent_log_context_offset = OFFSET(log_ctx),
 131     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
 132 };
 133
 134 unsigned swresample_version(void)
 135 {
 136     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
 137     return LIBSWRESAMPLE_VERSION_INT;
 138 }
 139
 140 const char *swresample_configuration(void)
 141 {
 142     return FFMPEG_CONFIGURATION;
 143 }
 144
 145 const char *swresample_license(void)
 146 {
 147 #define LICENSE_PREFIX "libswresample license: "
 148     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
 149 }
 150
 151 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
 152     if(!s || s->in_convert) // s needs to be allocated but not initialized
 153         return AVERROR(EINVAL);
 154     s->channel_map = channel_map;
 155     return 0;
 156 }
 157
 158 const AVClass *swr_get_class(void)
 159 {
 160     return &av_class;
 161 }
 162
 163 av_cold struct SwrContext *swr_alloc(void){
 164     SwrContext *s= av_mallocz(sizeof(SwrContext));
 165     if(s){
 166         s->av_class= &av_class;
 167         av_opt_set_defaults(s);
 168     }
 169     return s;
 170 }
 171
 172 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
 173                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
 174                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
 175                                       int log_offset, void *log_ctx){
 176     if(!s) s= swr_alloc();
 177     if(!s) return NULL;
 178
 179     s->log_level_offset= log_offset;
 180     s->log_ctx= log_ctx;
 181
 182     av_opt_set_int(s, "ocl", out_ch_layout,   0);
 183     av_opt_set_int(s, "osf", out_sample_fmt,  0);
 184     av_opt_set_int(s, "osr", out_sample_rate, 0);
 185     av_opt_set_int(s, "icl", in_ch_layout,    0);
 186     av_opt_set_int(s, "isf", in_sample_fmt,   0);
 187     av_opt_set_int(s, "isr", in_sample_rate,  0);
 188     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
 189     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
 190     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
 191     av_opt_set_int(s, "uch", 0, 0);
 192     return s;
 193 }
 194
 195 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 196     a->fmt   = fmt;
 197     a->bps   = av_get_bytes_per_sample(fmt);
 198     a->planar= av_sample_fmt_is_planar(fmt);
 199 }
 200
 201 static void free_temp(AudioData *a){
 202     av_free(a->data);
 203     memset(a, 0, sizeof(*a));
 204 }
 205
 206 av_cold void swr_free(SwrContext **ss){
 207     SwrContext *s= *ss;
 208     if(s){
 209         free_temp(&s->postin);
 210         free_temp(&s->midbuf);
 211         free_temp(&s->preout);
 212         free_temp(&s->in_buffer);
 213         free_temp(&s->dither);
 214         swri_audio_convert_free(&s-> in_convert);
 215         swri_audio_convert_free(&s->out_convert);
 216         swri_audio_convert_free(&s->full_convert);
 217         if (s->resampler)
 218             s->resampler->free(&s->resample);
 219         swri_rematrix_free(s);
 220     }
 221
 222     av_freep(ss);
 223 }
 224
 225 av_cold int swr_init(struct SwrContext *s){
 226     s->in_buffer_index= 0;
 227     s->in_buffer_count= 0;
 228     s->resample_in_constraint= 0;
 229     free_temp(&s->postin);
 230     free_temp(&s->midbuf);
 231     free_temp(&s->preout);
 232     free_temp(&s->in_buffer);
 233     free_temp(&s->dither);
 234     memset(s->in.ch, 0, sizeof(s->in.ch));
 235     memset(s->out.ch, 0, sizeof(s->out.ch));
 236     swri_audio_convert_free(&s-> in_convert);
 237     swri_audio_convert_free(&s->out_convert);
 238     swri_audio_convert_free(&s->full_convert);
 239     swri_rematrix_free(s);
 240
 241     s->flushed = 0;
 242
 243     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 244         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 245         return AVERROR(EINVAL);
 246     }
 247     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 248         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 249         return AVERROR(EINVAL);
 250     }
 251
 252     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 253         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 254             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 255         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 256             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 257         }else{
 258             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 259             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 260         }
 261     }
 262
 263     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 264         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 265         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 266         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 267         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));
 268         return AVERROR(EINVAL);
 269     }
 270
 271     switch(s->engine){
 272 #if CONFIG_LIBSOXR
 273         extern struct Resampler const soxr_resampler;
 274         case SWR_ENGINE_SOXR: s->resampler = &soxr_resampler; break;
 275 #endif
 276         case SWR_ENGINE_SWR : s->resampler = &swri_resampler; break;
 277         default:
 278             av_log(s, AV_LOG_ERROR, "Requested resampling engine is unavailable\n");
 279             return AVERROR(EINVAL);
 280     }
 281
 282     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 283     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 284
 285     if (s->async) {
 286         if (s->min_compensation >= FLT_MAX/2)
 287             s->min_compensation = 0.001;
 288         if (s->async > 1.0001) {
 289             s->max_soft_compensation = s->async / (double) s->in_sample_rate;
 290         }
 291     }
 292
 293     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 294         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);
 295     }else
 296         s->resampler->free(&s->resample);
 297     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 298         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 299         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 300         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 301         && s->resample){
 302         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 303         return -1;
 304     }
 305
 306     if(!s->used_ch_count)
 307         s->used_ch_count= s->in.ch_count;
 308
 309     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 310         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 311         s-> in_ch_layout= 0;
 312     }
 313
 314     if(!s-> in_ch_layout)
 315         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 316     if(!s->out_ch_layout)
 317         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 318
 319     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 320                  s->rematrix_custom;
 321
 322 #define RSC 1 //FIXME finetune
 323     if(!s-> in.ch_count)
 324         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 325     if(!s->used_ch_count)
 326         s->used_ch_count= s->in.ch_count;
 327     if(!s->out.ch_count)
 328         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 329
 330     if(!s-> in.ch_count){
 331         av_assert0(!s->in_ch_layout);
 332         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 333         return -1;
 334     }
 335
 336     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 337         char l1[1024], l2[1024];
 338         av_get_channel_layout_string(l1, sizeof(l1), s-> in.ch_count, s-> in_ch_layout);
 339         av_get_channel_layout_string(l2, sizeof(l2), s->out.ch_count, s->out_ch_layout);
 340         av_log(s, AV_LOG_ERROR, "Rematrix is needed between %s and %s "
 341                "but there is not enough information to do it\n", l1, l2);
 342         return -1;
 343     }
 344
 345 av_assert0(s->used_ch_count);
 346 av_assert0(s->out.ch_count);
 347     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 348
 349     s->in_buffer= s->in;
 350
 351     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
 352         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 353                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 354         return 0;
 355     }
 356
 357     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 358                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 359     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 360                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 361
 362
 363     s->postin= s->in;
 364     s->preout= s->out;
 365     s->midbuf= s->in;
 366
 367     if(s->channel_map){
 368         s->postin.ch_count=
 369         s->midbuf.ch_count= s->used_ch_count;
 370         if(s->resample)
 371             s->in_buffer.ch_count= s->used_ch_count;
 372     }
 373     if(!s->resample_first){
 374         s->midbuf.ch_count= s->out.ch_count;
 375         if(s->resample)
 376             s->in_buffer.ch_count = s->out.ch_count;
 377     }
 378
 379     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 380     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 381     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 382
 383     if(s->resample){
 384         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 385     }
 386
 387     s->dither = s->preout;
 388
 389     if(s->rematrix || s->dither_method)
 390         return swri_rematrix_init(s);
 391
 392     return 0;
 393 }
 394
 395 int swri_realloc_audio(AudioData *a, int count){
 396     int i, countb;
 397     AudioData old;
 398
 399     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 400         return AVERROR(EINVAL);
 401
 402     if(a->count >= count)
 403         return 0;
 404
 405     count*=2;
 406
 407     countb= FFALIGN(count*a->bps, ALIGN);
 408     old= *a;
 409
 410     av_assert0(a->bps);
 411     av_assert0(a->ch_count);
 412
 413     a->data= av_mallocz(countb*a->ch_count);
 414     if(!a->data)
 415         return AVERROR(ENOMEM);
 416     for(i=0; i<a->ch_count; i++){
 417         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 418         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 419     }
 420     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 421     av_free(old.data);
 422     a->count= count;
 423
 424     return 1;
 425 }
 426
 427 static void copy(AudioData *out, AudioData *in,
 428                  int count){
 429     av_assert0(out->planar == in->planar);
 430     av_assert0(out->bps == in->bps);
 431     av_assert0(out->ch_count == in->ch_count);
 432     if(out->planar){
 433         int ch;
 434         for(ch=0; ch<out->ch_count; ch++)
 435             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 436     }else
 437         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 438 }
 439
 440 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 441     int i;
 442     if(!in_arg){
 443         memset(out->ch, 0, sizeof(out->ch));
 444     }else if(out->planar){
 445         for(i=0; i<out->ch_count; i++)
 446             out->ch[i]= in_arg[i];
 447     }else{
 448         for(i=0; i<out->ch_count; i++)
 449             out->ch[i]= in_arg[0] + i*out->bps;
 450     }
 451 }
 452
 453 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 454     int i;
 455     if(out->planar){
 456         for(i=0; i<out->ch_count; i++)
 457             in_arg[i]= out->ch[i];
 458     }else{
 459         in_arg[0]= out->ch[0];
 460     }
 461 }
 462
 463 /**
 464  *
 465  * out may be equal in.
 466  */
 467 static void buf_set(AudioData *out, AudioData *in, int count){
 468     int ch;
 469     if(in->planar){
 470         for(ch=0; ch<out->ch_count; ch++)
 471             out->ch[ch]= in->ch[ch] + count*out->bps;
 472     }else{
 473         for(ch=out->ch_count-1; ch>=0; ch--)
 474             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 475     }
 476 }
 477
 478 /**
 479  *
 480  * @return number of samples output per channel
 481  */
 482 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 483                              const AudioData * in_param, int in_count){
 484     AudioData in, out, tmp;
 485     int ret_sum=0;
 486     int border=0;
 487
 488     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 489     av_assert1(s->in_buffer.planar   == in_param->planar);
 490     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 491
 492     tmp=out=*out_param;
 493     in =  *in_param;
 494
 495     do{
 496         int ret, size, consumed;
 497         if(!s->resample_in_constraint && s->in_buffer_count){
 498             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 499             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 500             out_count -= ret;
 501             ret_sum += ret;
 502             buf_set(&out, &out, ret);
 503             s->in_buffer_count -= consumed;
 504             s->in_buffer_index += consumed;
 505
 506             if(!in_count)
 507                 break;
 508             if(s->in_buffer_count <= border){
 509                 buf_set(&in, &in, -s->in_buffer_count);
 510                 in_count += s->in_buffer_count;
 511                 s->in_buffer_count=0;
 512                 s->in_buffer_index=0;
 513                 border = 0;
 514             }
 515         }
 516
 517         if((s->flushed || in_count) && !s->in_buffer_count){
 518             s->in_buffer_index=0;
 519             ret= s->resampler->multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 520             out_count -= ret;
 521             ret_sum += ret;
 522             buf_set(&out, &out, ret);
 523             in_count -= consumed;
 524             buf_set(&in, &in, consumed);
 525         }
 526
 527         //TODO is this check sane considering the advanced copy avoidance below
 528         size= s->in_buffer_index + s->in_buffer_count + in_count;
 529         if(   size > s->in_buffer.count
 530            && s->in_buffer_count + in_count <= s->in_buffer_index){
 531             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 532             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 533             s->in_buffer_index=0;
 534         }else
 535             if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 536                 return ret;
 537
 538         if(in_count){
 539             int count= in_count;
 540             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 541
 542             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 543             copy(&tmp, &in, /*in_*/count);
 544             s->in_buffer_count += count;
 545             in_count -= count;
 546             border += count;
 547             buf_set(&in, &in, count);
 548             s->resample_in_constraint= 0;
 549             if(s->in_buffer_count != count || in_count)
 550                 continue;
 551         }
 552         break;
 553     }while(1);
 554
 555     s->resample_in_constraint= !!out_count;
 556
 557     return ret_sum;
 558 }
 559
 560 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 561                                                       AudioData *in , int  in_count){
 562     AudioData *postin, *midbuf, *preout;
 563     int ret/*, in_max*/;
 564     AudioData preout_tmp, midbuf_tmp;
 565
 566     if(s->full_convert){
 567         av_assert0(!s->resample);
 568         swri_audio_convert(s->full_convert, out, in, in_count);
 569         return out_count;
 570     }
 571
 572 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 573 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 574
 575     if((ret=swri_realloc_audio(&s->postin, in_count))<0)
 576         return ret;
 577     if(s->resample_first){
 578         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 579         if((ret=swri_realloc_audio(&s->midbuf, out_count))<0)
 580             return ret;
 581     }else{
 582         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 583         if((ret=swri_realloc_audio(&s->midbuf,  in_count))<0)
 584             return ret;
 585     }
 586     if((ret=swri_realloc_audio(&s->preout, out_count))<0)
 587         return ret;
 588
 589     postin= &s->postin;
 590
 591     midbuf_tmp= s->midbuf;
 592     midbuf= &midbuf_tmp;
 593     preout_tmp= s->preout;
 594     preout= &preout_tmp;
 595
 596     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 597         postin= in;
 598
 599     if(s->resample_first ? !s->resample : !s->rematrix)
 600         midbuf= postin;
 601
 602     if(s->resample_first ? !s->rematrix : !s->resample)
 603         preout= midbuf;
 604
 605     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 606         if(preout==in){
 607             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 608             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 609             copy(out, in, out_count);
 610             return out_count;
 611         }
 612         else if(preout==postin) preout= midbuf= postin= out;
 613         else if(preout==midbuf) preout= midbuf= out;
 614         else                    preout= out;
 615     }
 616
 617     if(in != postin){
 618         swri_audio_convert(s->in_convert, postin, in, in_count);
 619     }
 620
 621     if(s->resample_first){
 622         if(postin != midbuf)
 623             out_count= resample(s, midbuf, out_count, postin, in_count);
 624         if(midbuf != preout)
 625             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 626     }else{
 627         if(postin != midbuf)
 628             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 629         if(midbuf != preout)
 630             out_count= resample(s, preout, out_count, midbuf, in_count);
 631     }
 632
 633     if(preout != out && out_count){
 634         if(s->dither_method){
 635             int ch;
 636             int dither_count= FFMAX(out_count, 1<<16);
 637             av_assert0(preout != in);
 638
 639             if((ret=swri_realloc_audio(&s->dither, dither_count))<0)
 640                 return ret;
 641             if(ret)
 642                 for(ch=0; ch<s->dither.ch_count; ch++)
 643                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
 644             av_assert0(s->dither.ch_count == preout->ch_count);
 645
 646             if(s->dither_pos + out_count > s->dither.count)
 647                 s->dither_pos = 0;
 648
 649             for(ch=0; ch<preout->ch_count; ch++)
 650                 s->mix_2_1_f(preout->ch[ch], preout->ch[ch], s->dither.ch[ch] + s->dither.bps * s->dither_pos, s->native_one, 0, 0, out_count);
 651
 652             s->dither_pos += out_count;
 653         }
 654 //FIXME packed doesnt need more than 1 chan here!
 655         swri_audio_convert(s->out_convert, out, preout, out_count);
 656     }
 657     return out_count;
 658 }
 659
 660 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 661                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 662     AudioData * in= &s->in;
 663     AudioData *out= &s->out;
 664
 665     if(s->drop_output > 0){
 666         int ret;
 667         AudioData tmp = s->out;
 668         uint8_t *tmp_arg[SWR_CH_MAX];
 669         tmp.count = 0;
 670         tmp.data  = NULL;
 671         if((ret=swri_realloc_audio(&tmp, s->drop_output))<0)
 672             return ret;
 673
 674         reversefill_audiodata(&tmp, tmp_arg);
 675         s->drop_output *= -1; //FIXME find a less hackish solution
 676         ret = swr_convert(s, tmp_arg, -s->drop_output, in_arg, in_count); //FIXME optimize but this is as good as never called so maybe it doesnt matter
 677         s->drop_output *= -1;
 678         if(ret>0)
 679             s->drop_output -= ret;
 680
 681         av_freep(&tmp.data);
 682         if(s->drop_output || !out_arg)
 683             return 0;
 684         in_count = 0;
 685     }
 686
 687     if(!in_arg){
 688         if(s->resample){
 689             if (!s->flushed)
 690                 s->resampler->flush(s);
 691             s->resample_in_constraint = 0;
 692             s->flushed = 1;
 693         }else if(!s->in_buffer_count){
 694             return 0;
 695         }
 696     }else
 697         fill_audiodata(in ,  (void*)in_arg);
 698
 699     fill_audiodata(out, out_arg);
 700
 701     if(s->resample){
 702         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 703         if(ret>0 && !s->drop_output)
 704             s->outpts += ret * (int64_t)s->in_sample_rate;
 705         return ret;
 706     }else{
 707         AudioData tmp= *in;
 708         int ret2=0;
 709         int ret, size;
 710         size = FFMIN(out_count, s->in_buffer_count);
 711         if(size){
 712             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 713             ret= swr_convert_internal(s, out, size, &tmp, size);
 714             if(ret<0)
 715                 return ret;
 716             ret2= ret;
 717             s->in_buffer_count -= ret;
 718             s->in_buffer_index += ret;
 719             buf_set(out, out, ret);
 720             out_count -= ret;
 721             if(!s->in_buffer_count)
 722                 s->in_buffer_index = 0;
 723         }
 724
 725         if(in_count){
 726             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 727
 728             if(in_count > out_count) { //FIXME move after swr_convert_internal
 729                 if(   size > s->in_buffer.count
 730                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 731                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 732                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 733                     s->in_buffer_index=0;
 734                 }else
 735                     if((ret=swri_realloc_audio(&s->in_buffer, size)) < 0)
 736                         return ret;
 737             }
 738
 739             if(out_count){
 740                 size = FFMIN(in_count, out_count);
 741                 ret= swr_convert_internal(s, out, size, in, size);
 742                 if(ret<0)
 743                     return ret;
 744                 buf_set(in, in, ret);
 745                 in_count -= ret;
 746                 ret2 += ret;
 747             }
 748             if(in_count){
 749                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 750                 copy(&tmp, in, in_count);
 751                 s->in_buffer_count += in_count;
 752             }
 753         }
 754         if(ret2>0 && !s->drop_output)
 755             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 756         return ret2;
 757     }
 758 }
 759
 760 int swr_drop_output(struct SwrContext *s, int count){
 761     s->drop_output += count;
 762
 763     if(s->drop_output <= 0)
 764         return 0;
 765
 766     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 767     return swr_convert(s, NULL, s->drop_output, NULL, 0);
 768 }
 769
 770 int swr_inject_silence(struct SwrContext *s, int count){
 771     int ret, i;
 772     AudioData silence = s->in;
 773     uint8_t *tmp_arg[SWR_CH_MAX];
 774
 775     if(count <= 0)
 776         return 0;
 777
 778     silence.count = 0;
 779     silence.data  = NULL;
 780     if((ret=swri_realloc_audio(&silence, count))<0)
 781         return ret;
 782
 783     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
 784         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
 785     } else
 786         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
 787
 788     reversefill_audiodata(&silence, tmp_arg);
 789     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 790     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 791     av_freep(&silence.data);
 792     return ret;
 793 }
 794
 795 int64_t swr_get_delay(struct SwrContext *s, int64_t base){
 796     if (s->resampler && s->resample){
 797         return s->resampler->get_delay(s, base);
 798     }else{
 799         return (s->in_buffer_count*base + (s->in_sample_rate>>1))/ s->in_sample_rate;
 800     }
 801 }
 802
 803 int swr_set_compensation(struct SwrContext *s, int sample_delta, int compensation_distance){
 804     int ret;
 805
 806     if (!s || compensation_distance < 0)
 807         return AVERROR(EINVAL);
 808     if (!compensation_distance && sample_delta)
 809         return AVERROR(EINVAL);
 810     if (!s->resample) {
 811         s->flags |= SWR_FLAG_RESAMPLE;
 812         ret = swr_init(s);
 813         if (ret < 0)
 814             return ret;
 815     }
 816     if (!s->resampler->set_compensation){
 817         return AVERROR(EINVAL);
 818     }else{
 819         return s->resampler->set_compensation(s->resample, sample_delta, compensation_distance);
 820     }
 821 }
 822
 823 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 824     if(pts == INT64_MIN)
 825         return s->outpts;
 826     if(s->min_compensation >= FLT_MAX) {
 827         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 828     } else {
 829         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
 830         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 831
 832         if(fabs(fdelta) > s->min_compensation) {
 833             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
 834                 int ret;
 835                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 836                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 837                 if(ret<0){
 838                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 839                 }
 840             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 841                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 842                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 843                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 844                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 845                 swr_set_compensation(s, comp, duration);
 846             }
 847         }
 848
 849         return s->outpts;
 850     }
 851 }