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/audioconvert.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"                  ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
  45 {"in_channel_count"     ,  "Input Channel Count"        , OFFSET( in.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
  46 {"och"                  , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
  47 {"out_channel_count"    , "Output Channel Count"        , OFFSET(out.ch_count   ), AV_OPT_TYPE_INT  , {.i64=2                     }, 0      , SWR_CH_MAX, PARAM},
  48 {"uch"                  ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  49 {"used_channel_count"   ,   "Used Channel Count"        , OFFSET(used_ch_count  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_CH_MAX, PARAM},
  50 {"isr"                  ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  51 {"in_sample_rate"       ,  "Input Sample Rate"          , OFFSET( in_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  52 {"osr"                  , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  53 {"out_sample_rate"      , "Output Sample Rate"          , OFFSET(out_sample_rate), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , INT_MAX   , PARAM},
  54 {"isf"                  ,    "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"        ,    "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"                  ,   "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"       ,   "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"                  , "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"  , "Internal Sample Format"      , OFFSET(int_sample_fmt ), AV_OPT_TYPE_INT  , {.i64=AV_SAMPLE_FMT_NONE    }, -1     , AV_SAMPLE_FMT_FLTP, PARAM},
  60 {"icl"                  ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  61 {"in_channel_layout"    ,   "Input Channel Layout"      , OFFSET( in_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  62 {"ocl"                  ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  63 {"out_channel_layout"   ,  "Output Channel Layout"      , OFFSET(out_ch_layout  ), AV_OPT_TYPE_INT64, {.i64=0                     }, 0      , INT64_MAX , PARAM, "channel_layout"},
  64 {"clev"                 ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  65 {"center_mix_level"     ,    "Center Mix Level"         , OFFSET(clev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  66 {"slev"                 , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  67 {"surround_mix_level"   , "Sourround Mix Level"         , OFFSET(slev           ), AV_OPT_TYPE_FLOAT, {.dbl=C_30DB                }, -32    , 32        , PARAM},
  68 {"lfe_mix_level"        , "LFE Mix Level"               , OFFSET(lfe_mix_level  ), AV_OPT_TYPE_FLOAT, {.dbl=0                     }, -32    , 32        , PARAM},
  69 {"rmvol"                , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
  70 {"rematrix_volume"      , "Rematrix Volume"             , OFFSET(rematrix_volume), AV_OPT_TYPE_FLOAT, {.dbl=1.0                   }, -1000  , 1000      , PARAM},
  71 {"flags"                , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
  72 {"swr_flags"            , NULL                          , OFFSET(flags          ), AV_OPT_TYPE_FLAGS, {.i64=0                     }, 0      , UINT_MAX  , PARAM, "flags"},
  73 {"res"                  , "Force Resampling"            , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_FLAG_RESAMPLE     }, INT_MIN, INT_MAX   , PARAM, "flags"},
  74 {"dither_scale"         , "Dither Scale"                , OFFSET(dither_scale   ), AV_OPT_TYPE_FLOAT, {.dbl=1                     }, 0      , INT_MAX   , PARAM},
  75 {"dither_method"        , "Dither Method"               , OFFSET(dither_method  ), AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , SWR_DITHER_NB-1, PARAM, "dither_method"},
  76 {"rectangular"          , "Rectangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_RECTANGULAR}, INT_MIN, INT_MAX   , PARAM, "dither_method"},
  77 {"triangular"           ,  "Triangular Dither"          , 0                      , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR }, INT_MIN, INT_MAX   , PARAM, "dither_method"},
  78 {"triangular_hp"        , "Triangular Dither With High Pass" , 0                 , AV_OPT_TYPE_CONST, {.i64=SWR_DITHER_TRIANGULAR_HIGHPASS }, INT_MIN, INT_MAX, PARAM, "dither_method"},
  79 {"filter_size"          , "Resampling Filter Size"      , OFFSET(filter_size)    , AV_OPT_TYPE_INT  , {.i64=16                    }, 0      , INT_MAX   , PARAM },
  80 {"phase_shift"          , "Resampling Phase Shift"      , OFFSET(phase_shift)    , AV_OPT_TYPE_INT  , {.i64=10                    }, 0      , 30        , PARAM },
  81 {"linear_interp"        , "Use Linear Interpolation"    , OFFSET(linear_interp)  , AV_OPT_TYPE_INT  , {.i64=0                     }, 0      , 1         , PARAM },
  82 {"cutoff"               , "Cutoff Frequency Ratio"      , OFFSET(cutoff)         , AV_OPT_TYPE_DOUBLE,{.dbl=0.8                   }, 0      , 1         , PARAM },
  83 {"min_comp"             , "Minimum difference between timestamps and audio data (in seconds) below which no timestamp compensation of either kind is applied"
  84                                                         , OFFSET(min_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=FLT_MAX               }, 0      , FLT_MAX   , PARAM },
  85 {"min_hard_comp"        , "Minimum difference between timestamps and audio data (in seconds) to trigger padding/trimming the data."
  86                                                    , OFFSET(min_hard_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0.1                   }, 0      , INT_MAX   , PARAM },
  87 {"comp_duration"        , "Duration (in seconds) over which data is stretched/squeezed to make it match the timestamps."
  88                                               , OFFSET(soft_compensation_duration),AV_OPT_TYPE_FLOAT ,{.dbl=1                     }, 0      , INT_MAX   , PARAM },
  89 {"max_soft_comp"        , "Maximum factor by which data is stretched/squeezed to make it match the timestamps."
  90                                                    , OFFSET(max_soft_compensation),AV_OPT_TYPE_FLOAT ,{.dbl=0                     }, INT_MIN, INT_MAX   , PARAM },
  91 { "matrix_encoding"     , "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" },
  92     { "none",  "None",               0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_NONE  }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
  93     { "dolby", "Dolby",              0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DOLBY }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
  94     { "dplii", "Dolby Pro Logic II", 0, AV_OPT_TYPE_CONST, { .i64 = AV_MATRIX_ENCODING_DPLII }, INT_MIN, INT_MAX, PARAM, "matrix_encoding" },
  95 { "filter_type"         , "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" },
  96     { "cubic"           , "Cubic"                       , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_CUBIC            }, INT_MIN, INT_MAX, PARAM, "filter_type" },
  97     { "blackman_nuttall", "Blackman Nuttall Windowed Sinc", 0                    , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_BLACKMAN_NUTTALL }, INT_MIN, INT_MAX, PARAM, "filter_type" },
  98     { "kaiser"          , "Kaiser Windowed Sinc"        , 0                      , AV_OPT_TYPE_CONST, { .i64 = SWR_FILTER_TYPE_KAISER           }, INT_MIN, INT_MAX, PARAM, "filter_type" },
  99 { "kaiser_beta"         , "Kaiser Window Beta"          ,OFFSET(kaiser_beta)     , AV_OPT_TYPE_INT  , {.i64=9                     }, 2      , 16        , PARAM },
 100
 101 {0}
 102 };
 103
 104 static const char* context_to_name(void* ptr) {
 105     return "SWR";
 106 }
 107
 108 static const AVClass av_class = {
 109     .class_name                = "SWResampler",
 110     .item_name                 = context_to_name,
 111     .option                    = options,
 112     .version                   = LIBAVUTIL_VERSION_INT,
 113     .log_level_offset_offset   = OFFSET(log_level_offset),
 114     .parent_log_context_offset = OFFSET(log_ctx),
 115     .category                  = AV_CLASS_CATEGORY_SWRESAMPLER,
 116 };
 117
 118 unsigned swresample_version(void)
 119 {
 120     av_assert0(LIBSWRESAMPLE_VERSION_MICRO >= 100);
 121     return LIBSWRESAMPLE_VERSION_INT;
 122 }
 123
 124 const char *swresample_configuration(void)
 125 {
 126     return FFMPEG_CONFIGURATION;
 127 }
 128
 129 const char *swresample_license(void)
 130 {
 131 #define LICENSE_PREFIX "libswresample license: "
 132     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
 133 }
 134
 135 int swr_set_channel_mapping(struct SwrContext *s, const int *channel_map){
 136     if(!s || s->in_convert) // s needs to be allocated but not initialized
 137         return AVERROR(EINVAL);
 138     s->channel_map = channel_map;
 139     return 0;
 140 }
 141
 142 const AVClass *swr_get_class(void)
 143 {
 144     return &av_class;
 145 }
 146
 147 av_cold struct SwrContext *swr_alloc(void){
 148     SwrContext *s= av_mallocz(sizeof(SwrContext));
 149     if(s){
 150         s->av_class= &av_class;
 151         av_opt_set_defaults(s);
 152     }
 153     return s;
 154 }
 155
 156 struct SwrContext *swr_alloc_set_opts(struct SwrContext *s,
 157                                       int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
 158                                       int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
 159                                       int log_offset, void *log_ctx){
 160     if(!s) s= swr_alloc();
 161     if(!s) return NULL;
 162
 163     s->log_level_offset= log_offset;
 164     s->log_ctx= log_ctx;
 165
 166     av_opt_set_int(s, "ocl", out_ch_layout,   0);
 167     av_opt_set_int(s, "osf", out_sample_fmt,  0);
 168     av_opt_set_int(s, "osr", out_sample_rate, 0);
 169     av_opt_set_int(s, "icl", in_ch_layout,    0);
 170     av_opt_set_int(s, "isf", in_sample_fmt,   0);
 171     av_opt_set_int(s, "isr", in_sample_rate,  0);
 172     av_opt_set_int(s, "tsf", AV_SAMPLE_FMT_NONE,   0);
 173     av_opt_set_int(s, "ich", av_get_channel_layout_nb_channels(s-> in_ch_layout), 0);
 174     av_opt_set_int(s, "och", av_get_channel_layout_nb_channels(s->out_ch_layout), 0);
 175     av_opt_set_int(s, "uch", 0, 0);
 176     return s;
 177 }
 178
 179 static void set_audiodata_fmt(AudioData *a, enum AVSampleFormat fmt){
 180     a->fmt   = fmt;
 181     a->bps   = av_get_bytes_per_sample(fmt);
 182     a->planar= av_sample_fmt_is_planar(fmt);
 183 }
 184
 185 static void free_temp(AudioData *a){
 186     av_free(a->data);
 187     memset(a, 0, sizeof(*a));
 188 }
 189
 190 av_cold void swr_free(SwrContext **ss){
 191     SwrContext *s= *ss;
 192     if(s){
 193         free_temp(&s->postin);
 194         free_temp(&s->midbuf);
 195         free_temp(&s->preout);
 196         free_temp(&s->in_buffer);
 197         free_temp(&s->dither);
 198         swri_audio_convert_free(&s-> in_convert);
 199         swri_audio_convert_free(&s->out_convert);
 200         swri_audio_convert_free(&s->full_convert);
 201         swri_resample_free(&s->resample);
 202         swri_rematrix_free(s);
 203     }
 204
 205     av_freep(ss);
 206 }
 207
 208 av_cold int swr_init(struct SwrContext *s){
 209     s->in_buffer_index= 0;
 210     s->in_buffer_count= 0;
 211     s->resample_in_constraint= 0;
 212     free_temp(&s->postin);
 213     free_temp(&s->midbuf);
 214     free_temp(&s->preout);
 215     free_temp(&s->in_buffer);
 216     free_temp(&s->dither);
 217     memset(s->in.ch, 0, sizeof(s->in.ch));
 218     memset(s->out.ch, 0, sizeof(s->out.ch));
 219     swri_audio_convert_free(&s-> in_convert);
 220     swri_audio_convert_free(&s->out_convert);
 221     swri_audio_convert_free(&s->full_convert);
 222     swri_rematrix_free(s);
 223
 224     s->flushed = 0;
 225
 226     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 227         av_log(s, AV_LOG_ERROR, "Requested input sample format %d is invalid\n", s->in_sample_fmt);
 228         return AVERROR(EINVAL);
 229     }
 230     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 231         av_log(s, AV_LOG_ERROR, "Requested output sample format %d is invalid\n", s->out_sample_fmt);
 232         return AVERROR(EINVAL);
 233     }
 234
 235     if(s->int_sample_fmt == AV_SAMPLE_FMT_NONE){
 236         if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_S16P){
 237             s->int_sample_fmt= AV_SAMPLE_FMT_S16P;
 238         }else if(av_get_planar_sample_fmt(s->in_sample_fmt) <= AV_SAMPLE_FMT_FLTP){
 239             s->int_sample_fmt= AV_SAMPLE_FMT_FLTP;
 240         }else{
 241             av_log(s, AV_LOG_DEBUG, "Using double precision mode\n");
 242             s->int_sample_fmt= AV_SAMPLE_FMT_DBLP;
 243         }
 244     }
 245
 246     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 247         &&s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 248         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 249         &&s->int_sample_fmt != AV_SAMPLE_FMT_DBLP){
 250         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));
 251         return AVERROR(EINVAL);
 252     }
 253
 254     set_audiodata_fmt(&s-> in, s-> in_sample_fmt);
 255     set_audiodata_fmt(&s->out, s->out_sample_fmt);
 256
 257     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 258         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);
 259     }else
 260         swri_resample_free(&s->resample);
 261     if(    s->int_sample_fmt != AV_SAMPLE_FMT_S16P
 262         && s->int_sample_fmt != AV_SAMPLE_FMT_S32P
 263         && s->int_sample_fmt != AV_SAMPLE_FMT_FLTP
 264         && s->int_sample_fmt != AV_SAMPLE_FMT_DBLP
 265         && s->resample){
 266         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16/s32/flt/dbl\n");
 267         return -1;
 268     }
 269
 270     if(!s->used_ch_count)
 271         s->used_ch_count= s->in.ch_count;
 272
 273     if(s->used_ch_count && s-> in_ch_layout && s->used_ch_count != av_get_channel_layout_nb_channels(s-> in_ch_layout)){
 274         av_log(s, AV_LOG_WARNING, "Input channel layout has a different number of channels than the number of used channels, ignoring layout\n");
 275         s-> in_ch_layout= 0;
 276     }
 277
 278     if(!s-> in_ch_layout)
 279         s-> in_ch_layout= av_get_default_channel_layout(s->used_ch_count);
 280     if(!s->out_ch_layout)
 281         s->out_ch_layout= av_get_default_channel_layout(s->out.ch_count);
 282
 283     s->rematrix= s->out_ch_layout  !=s->in_ch_layout || s->rematrix_volume!=1.0 ||
 284                  s->rematrix_custom;
 285
 286 #define RSC 1 //FIXME finetune
 287     if(!s-> in.ch_count)
 288         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 289     if(!s->used_ch_count)
 290         s->used_ch_count= s->in.ch_count;
 291     if(!s->out.ch_count)
 292         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 293
 294     if(!s-> in.ch_count){
 295         av_assert0(!s->in_ch_layout);
 296         av_log(s, AV_LOG_ERROR, "Input channel count and layout are unset\n");
 297         return -1;
 298     }
 299
 300     if ((!s->out_ch_layout || !s->in_ch_layout) && s->used_ch_count != s->out.ch_count && !s->rematrix_custom) {
 301         av_log(s, AV_LOG_ERROR, "Rematrix is needed but there is not enough information to do it\n");
 302         return -1;
 303     }
 304
 305 av_assert0(s->used_ch_count);
 306 av_assert0(s->out.ch_count);
 307     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 308
 309     s->in_buffer= s->in;
 310
 311     if(!s->resample && !s->rematrix && !s->channel_map && !s->dither_method){
 312         s->full_convert = swri_audio_convert_alloc(s->out_sample_fmt,
 313                                                    s-> in_sample_fmt, s-> in.ch_count, NULL, 0);
 314         return 0;
 315     }
 316
 317     s->in_convert = swri_audio_convert_alloc(s->int_sample_fmt,
 318                                              s-> in_sample_fmt, s->used_ch_count, s->channel_map, 0);
 319     s->out_convert= swri_audio_convert_alloc(s->out_sample_fmt,
 320                                              s->int_sample_fmt, s->out.ch_count, NULL, 0);
 321
 322
 323     s->postin= s->in;
 324     s->preout= s->out;
 325     s->midbuf= s->in;
 326
 327     if(s->channel_map){
 328         s->postin.ch_count=
 329         s->midbuf.ch_count= s->used_ch_count;
 330         if(s->resample)
 331             s->in_buffer.ch_count= s->used_ch_count;
 332     }
 333     if(!s->resample_first){
 334         s->midbuf.ch_count= s->out.ch_count;
 335         if(s->resample)
 336             s->in_buffer.ch_count = s->out.ch_count;
 337     }
 338
 339     set_audiodata_fmt(&s->postin, s->int_sample_fmt);
 340     set_audiodata_fmt(&s->midbuf, s->int_sample_fmt);
 341     set_audiodata_fmt(&s->preout, s->int_sample_fmt);
 342
 343     if(s->resample){
 344         set_audiodata_fmt(&s->in_buffer, s->int_sample_fmt);
 345     }
 346
 347     s->dither = s->preout;
 348
 349     if(s->rematrix || s->dither_method)
 350         return swri_rematrix_init(s);
 351
 352     return 0;
 353 }
 354
 355 static int realloc_audio(AudioData *a, int count){
 356     int i, countb;
 357     AudioData old;
 358
 359     if(count < 0 || count > INT_MAX/2/a->bps/a->ch_count)
 360         return AVERROR(EINVAL);
 361
 362     if(a->count >= count)
 363         return 0;
 364
 365     count*=2;
 366
 367     countb= FFALIGN(count*a->bps, ALIGN);
 368     old= *a;
 369
 370     av_assert0(a->bps);
 371     av_assert0(a->ch_count);
 372
 373     a->data= av_mallocz(countb*a->ch_count);
 374     if(!a->data)
 375         return AVERROR(ENOMEM);
 376     for(i=0; i<a->ch_count; i++){
 377         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 378         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 379     }
 380     if(!a->planar) memcpy(a->ch[0], old.ch[0], a->count*a->ch_count*a->bps);
 381     av_free(old.data);
 382     a->count= count;
 383
 384     return 1;
 385 }
 386
 387 static void copy(AudioData *out, AudioData *in,
 388                  int count){
 389     av_assert0(out->planar == in->planar);
 390     av_assert0(out->bps == in->bps);
 391     av_assert0(out->ch_count == in->ch_count);
 392     if(out->planar){
 393         int ch;
 394         for(ch=0; ch<out->ch_count; ch++)
 395             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 396     }else
 397         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 398 }
 399
 400 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 401     int i;
 402     if(!in_arg){
 403         memset(out->ch, 0, sizeof(out->ch));
 404     }else if(out->planar){
 405         for(i=0; i<out->ch_count; i++)
 406             out->ch[i]= in_arg[i];
 407     }else{
 408         for(i=0; i<out->ch_count; i++)
 409             out->ch[i]= in_arg[0] + i*out->bps;
 410     }
 411 }
 412
 413 static void reversefill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 414     int i;
 415     if(out->planar){
 416         for(i=0; i<out->ch_count; i++)
 417             in_arg[i]= out->ch[i];
 418     }else{
 419         in_arg[0]= out->ch[0];
 420     }
 421 }
 422
 423 /**
 424  *
 425  * out may be equal in.
 426  */
 427 static void buf_set(AudioData *out, AudioData *in, int count){
 428     int ch;
 429     if(in->planar){
 430         for(ch=0; ch<out->ch_count; ch++)
 431             out->ch[ch]= in->ch[ch] + count*out->bps;
 432     }else{
 433         for(ch=out->ch_count-1; ch>=0; ch--)
 434             out->ch[ch]= in->ch[0] + (ch + count*out->ch_count) * out->bps;
 435     }
 436 }
 437
 438 /**
 439  *
 440  * @return number of samples output per channel
 441  */
 442 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 443                              const AudioData * in_param, int in_count){
 444     AudioData in, out, tmp;
 445     int ret_sum=0;
 446     int border=0;
 447
 448     av_assert1(s->in_buffer.ch_count == in_param->ch_count);
 449     av_assert1(s->in_buffer.planar   == in_param->planar);
 450     av_assert1(s->in_buffer.fmt      == in_param->fmt);
 451
 452     tmp=out=*out_param;
 453     in =  *in_param;
 454
 455     do{
 456         int ret, size, consumed;
 457         if(!s->resample_in_constraint && s->in_buffer_count){
 458             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 459             ret= swri_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 460             out_count -= ret;
 461             ret_sum += ret;
 462             buf_set(&out, &out, ret);
 463             s->in_buffer_count -= consumed;
 464             s->in_buffer_index += consumed;
 465
 466             if(!in_count)
 467                 break;
 468             if(s->in_buffer_count <= border){
 469                 buf_set(&in, &in, -s->in_buffer_count);
 470                 in_count += s->in_buffer_count;
 471                 s->in_buffer_count=0;
 472                 s->in_buffer_index=0;
 473                 border = 0;
 474             }
 475         }
 476
 477         if(in_count && !s->in_buffer_count){
 478             s->in_buffer_index=0;
 479             ret= swri_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 480             out_count -= ret;
 481             ret_sum += ret;
 482             buf_set(&out, &out, ret);
 483             in_count -= consumed;
 484             buf_set(&in, &in, consumed);
 485         }
 486
 487         //TODO is this check sane considering the advanced copy avoidance below
 488         size= s->in_buffer_index + s->in_buffer_count + in_count;
 489         if(   size > s->in_buffer.count
 490            && s->in_buffer_count + in_count <= s->in_buffer_index){
 491             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 492             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 493             s->in_buffer_index=0;
 494         }else
 495             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 496                 return ret;
 497
 498         if(in_count){
 499             int count= in_count;
 500             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 501
 502             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 503             copy(&tmp, &in, /*in_*/count);
 504             s->in_buffer_count += count;
 505             in_count -= count;
 506             border += count;
 507             buf_set(&in, &in, count);
 508             s->resample_in_constraint= 0;
 509             if(s->in_buffer_count != count || in_count)
 510                 continue;
 511         }
 512         break;
 513     }while(1);
 514
 515     s->resample_in_constraint= !!out_count;
 516
 517     return ret_sum;
 518 }
 519
 520 static int swr_convert_internal(struct SwrContext *s, AudioData *out, int out_count,
 521                                                       AudioData *in , int  in_count){
 522     AudioData *postin, *midbuf, *preout;
 523     int ret/*, in_max*/;
 524     AudioData preout_tmp, midbuf_tmp;
 525
 526     if(s->full_convert){
 527         av_assert0(!s->resample);
 528         swri_audio_convert(s->full_convert, out, in, in_count);
 529         return out_count;
 530     }
 531
 532 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 533 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 534
 535     if((ret=realloc_audio(&s->postin, in_count))<0)
 536         return ret;
 537     if(s->resample_first){
 538         av_assert0(s->midbuf.ch_count == s->used_ch_count);
 539         if((ret=realloc_audio(&s->midbuf, out_count))<0)
 540             return ret;
 541     }else{
 542         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 543         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
 544             return ret;
 545     }
 546     if((ret=realloc_audio(&s->preout, out_count))<0)
 547         return ret;
 548
 549     postin= &s->postin;
 550
 551     midbuf_tmp= s->midbuf;
 552     midbuf= &midbuf_tmp;
 553     preout_tmp= s->preout;
 554     preout= &preout_tmp;
 555
 556     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar && !s->channel_map)
 557         postin= in;
 558
 559     if(s->resample_first ? !s->resample : !s->rematrix)
 560         midbuf= postin;
 561
 562     if(s->resample_first ? !s->rematrix : !s->resample)
 563         preout= midbuf;
 564
 565     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 566         if(preout==in){
 567             out_count= FFMIN(out_count, in_count); //TODO check at the end if this is needed or redundant
 568             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 569             copy(out, in, out_count);
 570             return out_count;
 571         }
 572         else if(preout==postin) preout= midbuf= postin= out;
 573         else if(preout==midbuf) preout= midbuf= out;
 574         else                    preout= out;
 575     }
 576
 577     if(in != postin){
 578         swri_audio_convert(s->in_convert, postin, in, in_count);
 579     }
 580
 581     if(s->resample_first){
 582         if(postin != midbuf)
 583             out_count= resample(s, midbuf, out_count, postin, in_count);
 584         if(midbuf != preout)
 585             swri_rematrix(s, preout, midbuf, out_count, preout==out);
 586     }else{
 587         if(postin != midbuf)
 588             swri_rematrix(s, midbuf, postin, in_count, midbuf==out);
 589         if(midbuf != preout)
 590             out_count= resample(s, preout, out_count, midbuf, in_count);
 591     }
 592
 593     if(preout != out && out_count){
 594         if(s->dither_method){
 595             int ch;
 596             int dither_count= FFMAX(out_count, 1<<16);
 597             av_assert0(preout != in);
 598
 599             if((ret=realloc_audio(&s->dither, dither_count))<0)
 600                 return ret;
 601             if(ret)
 602                 for(ch=0; ch<s->dither.ch_count; ch++)
 603                     swri_get_dither(s, s->dither.ch[ch], s->dither.count, 12345678913579<<ch, s->out_sample_fmt, s->int_sample_fmt);
 604             av_assert0(s->dither.ch_count == preout->ch_count);
 605
 606             if(s->dither_pos + out_count > s->dither.count)
 607                 s->dither_pos = 0;
 608
 609             for(ch=0; ch<preout->ch_count; ch++)
 610                 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);
 611
 612             s->dither_pos += out_count;
 613         }
 614 //FIXME packed doesnt need more than 1 chan here!
 615         swri_audio_convert(s->out_convert, out, preout, out_count);
 616     }
 617     return out_count;
 618 }
 619
 620 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 621                                 const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 622     AudioData * in= &s->in;
 623     AudioData *out= &s->out;
 624
 625     if(s->drop_output > 0){
 626         int ret;
 627         AudioData tmp = s->out;
 628         uint8_t *tmp_arg[SWR_CH_MAX];
 629         tmp.count = 0;
 630         tmp.data  = NULL;
 631         if((ret=realloc_audio(&tmp, s->drop_output))<0)
 632             return ret;
 633
 634         reversefill_audiodata(&tmp, tmp_arg);
 635         s->drop_output *= -1; //FIXME find a less hackish solution
 636         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
 637         s->drop_output *= -1;
 638         if(ret>0)
 639             s->drop_output -= ret;
 640
 641         av_freep(&tmp.data);
 642         if(s->drop_output || !out_arg)
 643             return 0;
 644         in_count = 0;
 645     }
 646
 647     if(!in_arg){
 648         if(s->in_buffer_count){
 649             if (s->resample && !s->flushed) {
 650                 AudioData *a= &s->in_buffer;
 651                 int i, j, ret;
 652                 if((ret=realloc_audio(a, s->in_buffer_index + 2*s->in_buffer_count)) < 0)
 653                     return ret;
 654                 av_assert0(a->planar);
 655                 for(i=0; i<a->ch_count; i++){
 656                     for(j=0; j<s->in_buffer_count; j++){
 657                         memcpy(a->ch[i] + (s->in_buffer_index+s->in_buffer_count+j  )*a->bps,
 658                             a->ch[i] + (s->in_buffer_index+s->in_buffer_count-j-1)*a->bps, a->bps);
 659                     }
 660                 }
 661                 s->in_buffer_count += (s->in_buffer_count+1)/2;
 662                 s->resample_in_constraint = 0;
 663                 s->flushed = 1;
 664             }
 665         }else{
 666             return 0;
 667         }
 668     }else
 669         fill_audiodata(in ,  (void*)in_arg);
 670
 671     fill_audiodata(out, out_arg);
 672
 673     if(s->resample){
 674         int ret = swr_convert_internal(s, out, out_count, in, in_count);
 675         if(ret>0 && !s->drop_output)
 676             s->outpts += ret * (int64_t)s->in_sample_rate;
 677         return ret;
 678     }else{
 679         AudioData tmp= *in;
 680         int ret2=0;
 681         int ret, size;
 682         size = FFMIN(out_count, s->in_buffer_count);
 683         if(size){
 684             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 685             ret= swr_convert_internal(s, out, size, &tmp, size);
 686             if(ret<0)
 687                 return ret;
 688             ret2= ret;
 689             s->in_buffer_count -= ret;
 690             s->in_buffer_index += ret;
 691             buf_set(out, out, ret);
 692             out_count -= ret;
 693             if(!s->in_buffer_count)
 694                 s->in_buffer_index = 0;
 695         }
 696
 697         if(in_count){
 698             size= s->in_buffer_index + s->in_buffer_count + in_count - out_count;
 699
 700             if(in_count > out_count) { //FIXME move after swr_convert_internal
 701                 if(   size > s->in_buffer.count
 702                 && s->in_buffer_count + in_count - out_count <= s->in_buffer_index){
 703                     buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 704                     copy(&s->in_buffer, &tmp, s->in_buffer_count);
 705                     s->in_buffer_index=0;
 706                 }else
 707                     if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 708                         return ret;
 709             }
 710
 711             if(out_count){
 712                 size = FFMIN(in_count, out_count);
 713                 ret= swr_convert_internal(s, out, size, in, size);
 714                 if(ret<0)
 715                     return ret;
 716                 buf_set(in, in, ret);
 717                 in_count -= ret;
 718                 ret2 += ret;
 719             }
 720             if(in_count){
 721                 buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 722                 copy(&tmp, in, in_count);
 723                 s->in_buffer_count += in_count;
 724             }
 725         }
 726         if(ret2>0 && !s->drop_output)
 727             s->outpts += ret2 * (int64_t)s->in_sample_rate;
 728         return ret2;
 729     }
 730 }
 731
 732 int swr_drop_output(struct SwrContext *s, int count){
 733     s->drop_output += count;
 734
 735     if(s->drop_output <= 0)
 736         return 0;
 737
 738     av_log(s, AV_LOG_VERBOSE, "discarding %d audio samples\n", count);
 739     return swr_convert(s, NULL, s->drop_output, NULL, 0);
 740 }
 741
 742 int swr_inject_silence(struct SwrContext *s, int count){
 743     int ret, i;
 744     AudioData silence = s->in;
 745     uint8_t *tmp_arg[SWR_CH_MAX];
 746
 747     if(count <= 0)
 748         return 0;
 749
 750     silence.count = 0;
 751     silence.data  = NULL;
 752     if((ret=realloc_audio(&silence, count))<0)
 753         return ret;
 754
 755     if(silence.planar) for(i=0; i<silence.ch_count; i++) {
 756         memset(silence.ch[i], silence.bps==1 ? 0x80 : 0, count*silence.bps);
 757     } else
 758         memset(silence.ch[0], silence.bps==1 ? 0x80 : 0, count*silence.bps*silence.ch_count);
 759
 760     reversefill_audiodata(&silence, tmp_arg);
 761     av_log(s, AV_LOG_VERBOSE, "adding %d audio samples of silence\n", count);
 762     ret = swr_convert(s, NULL, 0, (const uint8_t**)tmp_arg, count);
 763     av_freep(&silence.data);
 764     return ret;
 765 }
 766
 767 int64_t swr_next_pts(struct SwrContext *s, int64_t pts){
 768     if(pts == INT64_MIN)
 769         return s->outpts;
 770     if(s->min_compensation >= FLT_MAX) {
 771         return (s->outpts = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate));
 772     } else {
 773         int64_t delta = pts - swr_get_delay(s, s->in_sample_rate * (int64_t)s->out_sample_rate) - s->outpts;
 774         double fdelta = delta /(double)(s->in_sample_rate * (int64_t)s->out_sample_rate);
 775
 776         if(fabs(fdelta) > s->min_compensation) {
 777             if(!s->outpts || fabs(fdelta) > s->min_hard_compensation){
 778                 int ret;
 779                 if(delta > 0) ret = swr_inject_silence(s,  delta / s->out_sample_rate);
 780                 else          ret = swr_drop_output   (s, -delta / s-> in_sample_rate);
 781                 if(ret<0){
 782                     av_log(s, AV_LOG_ERROR, "Failed to compensate for timestamp delta of %f\n", fdelta);
 783                 }
 784             } else if(s->soft_compensation_duration && s->max_soft_compensation) {
 785                 int duration = s->out_sample_rate * s->soft_compensation_duration;
 786                 double max_soft_compensation = s->max_soft_compensation / (s->max_soft_compensation < 0 ? -s->in_sample_rate : 1);
 787                 int comp = av_clipf(fdelta, -max_soft_compensation, max_soft_compensation) * duration ;
 788                 av_log(s, AV_LOG_VERBOSE, "compensating audio timestamp drift:%f compensation:%d in:%d\n", fdelta, comp, duration);
 789                 swr_set_compensation(s, comp, duration);
 790             }
 791         }
 792
 793         return s->outpts;
 794     }
 795 }