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