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