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_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  55 {"in_sample_fmt"        , "set input sample format"     , OFFSET( in_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  56 {"osf"                  , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  57 {"out_sample_fmt"       , "set output sample format"    , OFFSET(out_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_NB-1+256, PARAM},
  58 {"tsf"                  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
  59 {"internal_sample_fmt"  , "set internal sample format"  , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, 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=0.8                   }, 0      , 1         , PARAM },
  87 {"min_comp"             , "set minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
  88                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
  89 {"min_hard_comp"        , "set minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
  90                                                         , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
  91 {"comp_duration"        , "set duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
  92                                                         , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
  93 {"max_soft_comp"        , "set maximum factor by which data is stretched/squeezed to make it match the timestamps."
  94                                                         , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
  95
  96 { "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" },
  97     { "none",  "select none",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
  98     { "dolby", "select Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
  99     { "dplii", "select Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
 100
 101 { "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" },
 102     { "cubic"           , "select cubic"                , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 103     { "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" },
 104     { "kaiser"          , "select Kaiser Windowed Sinc" , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
 105
 106 { "kaiser_beta"         , "set Kaiser Window Beta"      , OFFSET(kaiser_beta)    , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
 107
 108 {0}
 109 };
 110
 111 static const char* context_to_name(void* ptr) {
 112     return "SWR";
 113 }
 114
 115 static const AVClass av_class = {
 116     .class_name                = "SWResampler",
 117     .item_name                 = context_to_name,
 118     .option                    = options,
 119     .version                   = LIBAVUTIL_VERSION_INT,
 120     .log_level_offset_offset   = OFFSET(log_level_offset),
 121     .parent_log_context_offset = OFFSET(log_ctx),
 122     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
 123 };
 124
 125 unsigned swresample_version(void)
 126 {
 127     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
 128     return LIBSWRESAMPLE_VERSION_INT;
 129 }
 130
 131 const char *swresample_configuration(void)
 132 {
 133     return FFMPEG_CONFIGURATION;
 134 }
 135
 136 const char *swresample_license(void)
 137 {
 138 #define LICENSE_PREFIX "libswresample license: "
 139     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
 140 }
 141
 142 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
 143     if(!s || s->in_convert) // s needs to be allocated but not initialized
 144         return AVERROR(EINVAL);
 145     s->channel_map = channel_map;
 146     return 0;
 147 }
 148
 149 const AVClass *swr_get_class(void)
 150 {
 151     return &av_class;
 152 }
 153
 154 av_cold struct SwrContext *swr_alloc(void){
 155     SwrContext *s= av_mallocz(sizeof(SwrContext));
 156     if(s){
 157         s->av_class= &av_class;
 158         av_opt_set_defaults(s);
 159     }
 160     return s;
 161 }
 162
 163 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
 164                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
 165                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
 166                                       int log_offset, void *log_ctx){
 167     if(!s) s= swr_alloc();
 168     if(!s) return NULL;
 169
 170     s->log_level_offset= log_offset;
 171     s->log_ctx= log_ctx;
 172
 173     av_opt_set_int(s, "ocl", out_ch_layout,   0);
 174     av_opt_set_int(s, "osf", out_sample_fmt,  0);
 175     av_opt_set_int(s, "osr", out_sample_rate, 0);
 176     av_opt_set_int(s, "icl", in_ch_layout,    0);
 177     av_opt_set_int(s, "isf", in_sample_fmt,   0);
 178     av_opt_set_int(s, "isr", in_sample_rate,  0);
 179     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
 180     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
 181     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
 182     av_opt_set_int(s, "uch", 0, 0);
 183     return s;
 184 }
 185
 186 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 187     a->fmt   = fmt;
 188     a->bps   = av_get_bytes_per_sample(fmt);
 189     a->planar= av_sample_fmt_is_planar(fmt);
 190 }
 191
 192 static void free_temp(AudioData *a){
 193     av_free(a->data);
 194     memset(a, 0, sizeof(*a));
 195 }
 196
 197 av_cold void swr_free(SwrContext **ss){
 198     SwrContext *s= *ss;
 199     if(s){
 200         free_temp(&s->postin);
 201         free_temp(&s->midbuf);
 202         free_temp(&s->preout);
 203         free_temp(&s->in_buffer);
 204         free_temp(&s->dither);
 205         swri_audio_convert_free(&s-> in_convert);
 206         swri_audio_convert_free(&s->out_convert);
 207         swri_audio_convert_free(&s->full_convert);
 208         swri_resample_free(&s->resample);
 209         swri_rematrix_free(s);
 210     }
 211
 212     av_freep(ss);
 213 }
 214
 215 av_cold int swr_init(struct SwrContext *s){
 216     s->in_buffer_index= 0;
 217     s->in_buffer_count= 0;
 218     s->resample_in_constraint= 0;
 219     free_temp(&s->postin);
 220     free_temp(&s->midbuf);
 221     free_temp(&s->preout);
 222     free_temp(&s->in_buffer);
 223     free_temp(&s->dither);
 224     memset(s->in.ch, 0, sizeof(s->in.ch));
 225     memset(s->out.ch, 0, sizeof(s->out.ch));
 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     swri_rematrix_free(s);
 230
 231     s->flushed = 0;
 232
 233     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 234         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 235         return AVERROR(EINVAL);
 236     }
 237     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 238         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 239         return AVERROR(EINVAL);
 240     }
 241
 242     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 243         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 244             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 245         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 246             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 247         }else{
 248             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 249             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 250         }
 251     }
 252
 253     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 254         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 255         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 256         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 257         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));
 258         return AVERROR(EINVAL);
 259     }
 260
 261     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 262     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 263
 264     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 265         s->resample = swri_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, s->filter_size, s->phase_shift, s->linear_interp, s->cutoff, s->int_sample_fmt, s->filter_type, s->kaiser_beta);
 266     }else
 267         swri_resample_free(&s->resample);
 268     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 269         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 270         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 271         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 272         && s->resample){
 273         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 274         return -1;
 275     }
 276
 277     if(!s->used_ch_count)
 278         s->used_ch_count= s->in.ch_count;
 279
 280     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 281         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 282         s-> in_ch_layout= 0;
 283     }
 284
 285     if(!s-> in_ch_layout)
 286         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 287     if(!s->out_ch_layout)
 288         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 289
 290     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 291                  s->rematrix_custom;
 292
 293 #define RSC 1 //FIXME finetune
 294     if(!s-> in.ch_count)
 295         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 296     if(!s->used_ch_count)
 297         s->used_ch_count= s->in.ch_count;
 298     if(!s->out.ch_count)
 299         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 300
 301     if(!s-> in.ch_count){
 302         av_assert0(!s->in_ch_layout);
 303         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 304         return -1;
 305     }
 306
 307     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 308         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
 309         return -1;
 310     }
 311
 312 av_assert0(s->used_ch_count);
 313 av_assert0(s->out.ch_count);
 314     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 315
 316     s->in_buffer= s->in;
 317
 318     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
 319         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 320                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 321         return 0;
 322     }
 323
 324     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 325                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 326     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 327                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 328
 329
 330     s->postin= s->in;
 331     s->preout= s->out;
 332     s->midbuf= s->in;
 333
 334     if(s->channel_map){
 335         s->postin.ch_count=
 336         s->midbuf.ch_count= s->used_ch_count;
 337         if(s->resample)
 338             s->in_buffer.ch_count= s->used_ch_count;
 339     }
 340     if(!s->resample_first){
 341         s->midbuf.ch_count= s->out.ch_count;
 342         if(s->resample)
 343             s->in_buffer.ch_count = s->out.ch_count;
 344     }
 345
 346     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 347     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 348     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 349
 350     if(s->resample){
 351         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 352     }
 353
 354     s->dither = s->preout;
 355
 356     if(s->rematrix || s->dither_method)
 357         return swri_rematrix_init(s);
 358
 359     return 0;
 360 }
 361
 362 static int realloc_audio(AudioData *a, int count){
 363     int i, countb;
 364     AudioData old;
 365
 366     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 367         return AVERROR(EINVAL);
 368
 369     if(a->count >= count)
 370         return 0;
 371
 372     count*=2;
 373
 374     countb= FFALIGN(count*a->bps, ALIGN);
 375     old= *a;
 376
 377     av_assert0(a->bps);
 378     av_assert0(a->ch_count);
 379
 380     a->data= av_mallocz(countb*a->ch_count);
 381     if(!a->data)
 382         return AVERROR(ENOMEM);
 383     for(i=0; i<a->ch_count; i++){
 384         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 385         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 386     }
 387     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 388     av_free(old.data);
 389     a->count= count;
 390
 391     return 1;
 392 }
 393
 394 static void copy(AudioData *out, AudioData *in,
 395                  int count){
 396     av_assert0(out->planar == in->planar);
 397     av_assert0(out->bps == in->bps);
 398     av_assert0(out->ch_count == in->ch_count);
 399     if(out->planar){
 400         int ch;
 401         for(ch=0; ch<out->ch_count; ch++)
 402             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 403     }else
 404         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 405 }
 406
 407 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 408     int i;
 409     if(!in_arg){
 410         memset(out->ch, 0, sizeof(out->ch));
 411     }else if(out->planar){
 412         for(i=0; i<out->ch_count; i++)
 413             out->ch[i]= in_arg[i];
 414     }else{
 415         for(i=0; i<out->ch_count; i++)
 416             out->ch[i]= in_arg[0] + i*out->bps;
 417     }
 418 }
 419
 420 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 421     int i;
 422     if(out->planar){
 423         for(i=0; i<out->ch_count; i++)
 424             in_arg[i]= out->ch[i];
 425     }else{
 426         in_arg[0]= out->ch[0];
 427     }
 428 }
 429
 430 /**
 431  *
 432  * out may be equal in.
 433  */
 434 static void buf_set(AudioData *out, AudioData *in, int count){
 435     int ch;
 436     if(in->planar){
 437         for(ch=0; ch<out->ch_count; ch++)
 438             out->ch[ch]= in->ch[ch] + count*out->bps;
 439     }else{
 440         for(ch=out->ch_count-1; ch>=0; ch--)
 441             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 442     }
 443 }
 444
 445 /**
 446  *
 447  * @return number of samples output per channel
 448  */
 449 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 450                              const AudioData * in_param, int in_count){
 451     AudioData in, out, tmp;
 452     int ret_sum=0;
 453     int border=0;
 454
 455     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 456     av_assert1(s->in_buffer.planar   == in_param->planar);
 457     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 458
 459     tmp=out=*out_param;
 460     in =  *in_param;
 461
 462     do{
 463         int ret, size, consumed;
 464         if(!s->resample_in_constraint && s->in_buffer_count){
 465             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 466             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 467             out_count -= ret;
 468             ret_sum += ret;
 469             buf_set(&out, &out, ret);
 470             s->in_buffer_count -= consumed;
 471             s->in_buffer_index += consumed;
 472
 473             if(!in_count)
 474                 break;
 475             if(s->in_buffer_count <= border){
 476                 buf_set(&in, &in, -s->in_buffer_count);
 477                 in_count += s->in_buffer_count;
 478                 s->in_buffer_count=0;
 479                 s->in_buffer_index=0;
 480                 border = 0;
 481             }
 482         }
 483
 484         if(in_count && !s->in_buffer_count){
 485             s->in_buffer_index=0;
 486             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 487             out_count -= ret;
 488             ret_sum += ret;
 489             buf_set(&out, &out, ret);
 490             in_count -= consumed;
 491             buf_set(&in, &in, consumed);
 492         }
 493
 494         //TODO is this check sane considering the advanced copy avoidance below
 495         size= s->in_buffer_index + s->in_buffer_count + in_count;
 496         if(   size > s->in_buffer.count
 497            && s->in_buffer_count + in_count <= s->in_buffer_index){
 498             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 499             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 500             s->in_buffer_index=0;
 501         }else
 502             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 503                 return ret;
 504
 505         if(in_count){
 506             int count= in_count;
 507             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 508
 509             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 510             copy(&tmp, &in, /*in_*/count);
 511             s->in_buffer_count += count;
 512             in_count -= count;
 513             border += count;
 514             buf_set(&in, &in, count);
 515             s->resample_in_constraint= 0;
 516             if(s->in_buffer_count != count || in_count)
 517                 continue;
 518         }
 519         break;
 520     }while(1);
 521
 522     s->resample_in_constraint= !!out_count;
 523
 524     return ret_sum;
 525 }
 526
 527 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 528                                                       AudioData *in , int  in_count){
 529     AudioData *postin, *midbuf, *preout;
 530     int ret/*, in_max*/;
 531     AudioData preout_tmp, midbuf_tmp;
 532
 533     if(s->full_convert){
 534         av_assert0(!s->resample);
 535         swri_audio_convert(s->full_convert, out, in, in_count);
 536         return out_count;
 537     }
 538
 539 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 540 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 541
 542     if((ret=realloc_audio(&s->postin, in_count))<0)
 543         return ret;
 544     if(s->resample_first){
 545         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 546         if((ret=realloc_audio(&s->midbuf, out_count))<0)
 547             return ret;
 548     }else{
 549         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 550         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
 551             return ret;
 552     }
 553     if((ret=realloc_audio(&s->preout, out_count))<0)
 554         return ret;
 555
 556     postin= &s->postin;
 557
 558     midbuf_tmp= s->midbuf;
 559     midbuf= &midbuf_tmp;
 560     preout_tmp= s->preout;
 561     preout= &preout_tmp;
 562
 563     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 564         postin= in;
 565
 566     if(s->resample_first ? !s->resample : !s->rematrix)
 567         midbuf= postin;
 568
 569     if(s->resample_first ? !s->rematrix : !s->resample)
 570         preout= midbuf;
 571
 572     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 573         if(preout==in){
 574             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 575             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 576             copy(out, in, out_count);
 577             return out_count;
 578         }
 579         else if(preout==postin) preout= midbuf= postin= out;
 580         else if(preout==midbuf) preout= midbuf= out;
 581         else                    preout= out;
 582     }
 583
 584     if(in != postin){
 585         swri_audio_convert(s->in_convert, postin, in, in_count);
 586     }
 587
 588     if(s->resample_first){
 589         if(postin != midbuf)
 590             out_count= resample(s, midbuf, out_count, postin, in_count);
 591         if(midbuf != preout)
 592             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 593     }else{
 594         if(postin != midbuf)
 595             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 596         if(midbuf != preout)
 597             out_count= resample(s, preout, out_count, midbuf, in_count);
 598     }
 599
 600     if(preout != out && out_count){
 601         if(s->dither_method){
 602             int ch;
 603             int dither_count= FFMAX(out_count, 1<<16);
 604             av_assert0(preout != in);
 605
 606             if((ret=realloc_audio(&s->dither, dither_count))<0)
 607                 return ret;
 608             if(ret)
 609                 for(ch=0; ch<s->dither.ch_count; ch++)
 610                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
 611             av_assert0(s->dither.ch_count == preout->ch_count);
 612
 613             if(s->dither_pos + out_count > s->dither.count)
 614                 s->dither_pos = 0;
 615
 616             for(ch=0; ch<preout->ch_count; ch++)
 617                 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);
 618
 619             s->dither_pos += out_count;
 620         }
 621 //FIXME packed doesnt need more than 1 chan here!
 622         swri_audio_convert(s->out_convert, out, preout, out_count);
 623     }
 624     return out_count;
 625 }
 626
 627 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 628                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 629     AudioData * in= &s->in;
 630     AudioData *out= &s->out;
 631
 632     if(s->drop_output > 0){
 633         int ret;
 634         AudioData tmp = s->out;
 635         uint8_t *tmp_arg[SWR_CH_MAX];
 636         tmp.count = 0;
 637         tmp.data  = NULL;
 638         if((ret=realloc_audio(&tmp, s->drop_output))<0)
 639             return ret;
 640
 641         reversefill_audiodata(&tmp, tmp_arg);
 642         s->drop_output *= -1; //FIXME find a less hackish solution
 643         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
 644         s->drop_output *= -1;
 645         if(ret>0)
 646             s->drop_output -= ret;
 647
 648         av_freep(&tmp.data);
 649         if(s->drop_output || !out_arg)
 650             return 0;
 651         in_count = 0;
 652     }
 653
 654     if(!in_arg){
 655         if(s->in_buffer_count){
 656             if (s->resample && !s->flushed) {
 657                 AudioData *a= &s->in_buffer;
 658                 int i, j, ret;
 659                 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
 660                     return ret;
 661                 av_assert0(a->planar);
 662                 for(i=0; i<a->ch_count; i++){
 663                     for(j=0; j<s->in_buffer_count; j++){
 664                         memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
 665                             a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
 666                     }
 667                 }
 668                 s->in_buffer_count += (s->in_buffer_count+1)/2;
 669                 s->resample_in_constraint = 0;
 670                 s->flushed = 1;
 671             }
 672         }else{
 673             return 0;
 674         }
 675     }else
 676         fill_audiodata(in ,  (void*)in_arg);
 677
 678     fill_audiodata(out, out_arg);
 679
 680     if(s->resample){
 681         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 682         if(ret>0 && !s->drop_output)
 683             s->outpts += ret * (int64_t)s->in_sample_rate;
 684         return ret;
 685     }else{
 686         AudioData tmp= *in;
 687         int ret2=0;
 688         int ret, size;
 689         size = FFMIN(out_count, s->in_buffer_count);
 690         if(size){
 691             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 692             ret= swr_convert_internal(s, out, size, &tmp, size);
 693             if(ret<0)
 694                 return ret;
 695             ret2= ret;
 696             s->in_buffer_count -= ret;
 697             s->in_buffer_index += ret;
 698             buf_set(out, out, ret);
 699             out_count -= ret;
 700             if(!s->in_buffer_count)
 701                 s->in_buffer_index = 0;
 702         }
 703
 704         if(in_count){
 705             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 706
 707             if(in_count > out_count) { //FIXME move after swr_convert_internal
 708                 if(   size > s->in_buffer.count
 709                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 710                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 711                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 712                     s->in_buffer_index=0;
 713                 }else
 714                     if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 715                         return ret;
 716             }
 717
 718             if(out_count){
 719                 size = FFMIN(in_count, out_count);
 720                 ret= swr_convert_internal(s, out, size, in, size);
 721                 if(ret<0)
 722                     return ret;
 723                 buf_set(in, in, ret);
 724                 in_count -= ret;
 725                 ret2 += ret;
 726             }
 727             if(in_count){
 728                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 729                 copy(&tmp, in, in_count);
 730                 s->in_buffer_count += in_count;
 731             }
 732         }
 733         if(ret2>0 && !s->drop_output)
 734             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 735         return ret2;
 736     }
 737 }
 738
 739 int swr_drop_output(struct SwrContext *s, int count){
 740     s->drop_output += count;
 741
 742     if(s->drop_output <= 0)
 743         return 0;
 744
 745     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 746     return swr_convert(s, NULL, s->drop_output, NULL, 0);
 747 }
 748
 749 int swr_inject_silence(struct SwrContext *s, int count){
 750     int ret, i;
 751     AudioData silence = s->in;
 752     uint8_t *tmp_arg[SWR_CH_MAX];
 753
 754     if(count <= 0)
 755         return 0;
 756
 757     silence.count = 0;
 758     silence.data  = NULL;
 759     if((ret=realloc_audio(&silence, count))<0)
 760         return ret;
 761
 762     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
 763         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
 764     } else
 765         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
 766
 767     reversefill_audiodata(&silence, tmp_arg);
 768     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 769     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 770     av_freep(&silence.data);
 771     return ret;
 772 }
 773
 774 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 775     if(pts == INT64_MIN)
 776         return s->outpts;
 777     if(s->min_compensation >= FLT_MAX) {
 778         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 779     } else {
 780         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
 781         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 782
 783         if(fabs(fdelta) > s->min_compensation) {
 784             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
 785                 int ret;
 786                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 787                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 788                 if(ret<0){
 789                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 790                 }
 791             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 792                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 793                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 794                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 795                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 796                 swr_set_compensation(s, comp, duration);
 797             }
 798         }
 799
 800         return s->outpts;
 801     }
 802 }