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

   1 /*
   2  * Copyright (C) 2011 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
  26 #define  C30DB  M_SQRT2
  27 #define  C15DB  1.189207115
  28 #define C__0DB  1.0
  29 #define C_15DB  0.840896415
  30 #define C_30DB  M_SQRT1_2
  31 #define C_45DB  0.594603558
  32 #define C_60DB  0.5
  33
  34
  35 //TODO split options array out?
  36 #define OFFSET(x) offsetof(SwrContext,x)
  37 static const AVOption options[]={
  38 {"ich",  "input channel count", OFFSET( in.ch_count   ), FF_OPT_TYPE_INT, {.dbl=2}, 1, SWR_CH_MAX, 0},
  39 {"och", "output channel count", OFFSET(out.ch_count   ), FF_OPT_TYPE_INT, {.dbl=2}, 1, SWR_CH_MAX, 0},
  40 {"isr",  "input sample rate"  , OFFSET( in_sample_rate), FF_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
  41 {"osr", "output sample rate"  , OFFSET(out_sample_rate), FF_OPT_TYPE_INT, {.dbl=48000}, 1, INT_MAX, 0},
  42 //{"ip" ,  "input planar"       , OFFSET( in.planar     ), FF_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
  43 //{"op" , "output planar"       , OFFSET(out.planar     ), FF_OPT_TYPE_INT, {.dbl=0},    0,       1, 0},
  44 {"isf",  "input sample format", OFFSET( in_sample_fmt ), FF_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
  45 {"osf", "output sample format", OFFSET(out_sample_fmt ), FF_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_S16}, 0, AV_SAMPLE_FMT_NB-1+256, 0},
  46 {"tsf", "internal sample format", OFFSET(int_sample_fmt ), FF_OPT_TYPE_INT, {.dbl=AV_SAMPLE_FMT_NONE}, -1, AV_SAMPLE_FMT_FLT, 0},
  47 {"icl",  "input channel layout" , OFFSET( in_ch_layout), FF_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
  48 {"ocl",  "output channel layout", OFFSET(out_ch_layout), FF_OPT_TYPE_INT64, {.dbl=0}, 0, INT64_MAX, 0, "channel_layout"},
  49 {"clev", "center mix level"     , OFFSET(clev)         , FF_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
  50 {"slev", "sourround mix level"  , OFFSET(slev)         , FF_OPT_TYPE_FLOAT, {.dbl=C_30DB}, 0, 4, 0},
  51 {"flags", NULL                  , OFFSET(flags)        , FF_OPT_TYPE_FLAGS, {.dbl=0}, 0,  UINT_MAX, 0, "flags"},
  52 {"res", "force resampling", 0, FF_OPT_TYPE_CONST, {.dbl=SWR_FLAG_RESAMPLE}, INT_MIN, INT_MAX, 0, "flags"},
  53
  54 {0}
  55 };
  56
  57 static const char* context_to_name(void* ptr) {
  58     return "SWR";
  59 }
  60
  61 static const AVClass av_class = { "SwrContext", context_to_name, options, LIBAVUTIL_VERSION_INT, OFFSET(log_level_offset), OFFSET(log_ctx) };
  62
  63 static int resample(SwrContext *s, AudioData *out_param, int out_count,
  64                              const AudioData * in_param, int in_count);
  65
  66 SwrContext *swr_alloc(void){
  67     SwrContext *s= av_mallocz(sizeof(SwrContext));
  68     if(s){
  69         s->av_class= &av_class;
  70         av_opt_set_defaults2(s, 0, 0);
  71     }
  72     return s;
  73 }
  74
  75 SwrContext *swr_alloc2(struct SwrContext *s, int64_t out_ch_layout, enum AVSampleFormat out_sample_fmt, int out_sample_rate,
  76                        int64_t  in_ch_layout, enum AVSampleFormat  in_sample_fmt, int  in_sample_rate,
  77                        int log_offset, void *log_ctx){
  78     if(!s) s= swr_alloc();
  79     if(!s) return NULL;
  80
  81     s->log_level_offset= log_offset;
  82     s->log_ctx= log_ctx;
  83
  84     av_set_int(s, "ocl", out_ch_layout);
  85     av_set_int(s, "osf", out_sample_fmt);
  86     av_set_int(s, "osr", out_sample_rate);
  87     av_set_int(s, "icl", in_ch_layout);
  88     av_set_int(s, "isf", in_sample_fmt);
  89     av_set_int(s, "isr", in_sample_rate);
  90
  91     s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
  92     s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
  93     s->int_sample_fmt = AV_SAMPLE_FMT_S16;
  94
  95     return s;
  96 }
  97
  98
  99 static void free_temp(AudioData *a){
 100     av_free(a->data);
 101     memset(a, 0, sizeof(*a));
 102 }
 103
 104 void swr_free(SwrContext **ss){
 105     SwrContext *s= *ss;
 106     if(s){
 107         free_temp(&s->postin);
 108         free_temp(&s->midbuf);
 109         free_temp(&s->preout);
 110         free_temp(&s->in_buffer);
 111         swr_audio_convert_free(&s-> in_convert);
 112         swr_audio_convert_free(&s->out_convert);
 113         swr_resample_free(&s->resample);
 114     }
 115
 116     av_freep(ss);
 117 }
 118
 119 static int64_t guess_layout(int ch){
 120     switch(ch){
 121     case 1: return AV_CH_LAYOUT_MONO;
 122     case 2: return AV_CH_LAYOUT_STEREO;
 123     case 5: return AV_CH_LAYOUT_5POINT0;
 124     case 6: return AV_CH_LAYOUT_5POINT1;
 125     case 7: return AV_CH_LAYOUT_7POINT0;
 126     case 8: return AV_CH_LAYOUT_7POINT1;
 127     default: return 0;
 128     }
 129 }
 130
 131 int swr_init(SwrContext *s){
 132     s->in_buffer_index= 0;
 133     s->in_buffer_count= 0;
 134     s->resample_in_constraint= 0;
 135     free_temp(&s->postin);
 136     free_temp(&s->midbuf);
 137     free_temp(&s->preout);
 138     free_temp(&s->in_buffer);
 139     swr_audio_convert_free(&s-> in_convert);
 140     swr_audio_convert_free(&s->out_convert);
 141
 142     s-> in.planar= s-> in_sample_fmt >= 0x100;
 143     s->out.planar= s->out_sample_fmt >= 0x100;
 144     s-> in_sample_fmt &= 0xFF;
 145     s->out_sample_fmt &= 0xFF;
 146
 147     if(s-> in_sample_fmt >= AV_SAMPLE_FMT_NB){
 148         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->in_sample_fmt));
 149         return AVERROR(EINVAL);
 150     }
 151     if(s->out_sample_fmt >= AV_SAMPLE_FMT_NB){
 152         av_log(s, AV_LOG_ERROR, "Requested sample format %s is invalid\n", av_get_sample_fmt_name(s->out_sample_fmt));
 153         return AVERROR(EINVAL);
 154     }
 155
 156     if(   s->int_sample_fmt != AV_SAMPLE_FMT_S16
 157         &&s->int_sample_fmt != AV_SAMPLE_FMT_FLT){
 158         av_log(s, AV_LOG_ERROR, "Requested sample format %s is not supported internally, only float & S16 is supported\n", av_get_sample_fmt_name(s->int_sample_fmt));
 159         return AVERROR(EINVAL);
 160     }
 161
 162     //FIXME should we allow/support using FLT on material that doesnt need it ?
 163     if(s->in_sample_fmt <= AV_SAMPLE_FMT_S16 || s->int_sample_fmt==AV_SAMPLE_FMT_S16){
 164         s->int_sample_fmt= AV_SAMPLE_FMT_S16;
 165     }else
 166         s->int_sample_fmt= AV_SAMPLE_FMT_FLT;
 167
 168
 169     if (s->out_sample_rate!=s->in_sample_rate || (s->flags & SWR_FLAG_RESAMPLE)){
 170         s->resample = swr_resample_init(s->resample, s->out_sample_rate, s->in_sample_rate, 16, 10, 0, 0.8);
 171     }else
 172         swr_resample_free(&s->resample);
 173     if(s->int_sample_fmt != AV_SAMPLE_FMT_S16 && s->resample){
 174         av_log(s, AV_LOG_ERROR, "Resampling only supported with internal s16 currently\n"); //FIXME
 175         return -1;
 176     }
 177
 178     if(!s-> in_ch_layout)
 179         s-> in_ch_layout= guess_layout(s->in.ch_count);
 180     if(!s->out_ch_layout)
 181         s->out_ch_layout= guess_layout(s->out.ch_count);
 182
 183     s->rematrix= s->out_ch_layout  !=s->in_ch_layout;
 184
 185 #define RSC 1 //FIXME finetune
 186     if(!s-> in.ch_count)
 187         s-> in.ch_count= av_get_channel_layout_nb_channels(s-> in_ch_layout);
 188     if(!s->out.ch_count)
 189         s->out.ch_count= av_get_channel_layout_nb_channels(s->out_ch_layout);
 190
 191 av_assert0(s-> in.ch_count);
 192 av_assert0(s->out.ch_count);
 193     s->resample_first= RSC*s->out.ch_count/s->in.ch_count - RSC < s->out_sample_rate/(float)s-> in_sample_rate - 1.0;
 194
 195     s-> in.bps= av_get_bits_per_sample_fmt(s-> in_sample_fmt)/8;
 196     s->int_bps= av_get_bits_per_sample_fmt(s->int_sample_fmt)/8;
 197     s->out.bps= av_get_bits_per_sample_fmt(s->out_sample_fmt)/8;
 198
 199     s->in_convert = swr_audio_convert_alloc(s->int_sample_fmt,
 200                                             s-> in_sample_fmt, s-> in.ch_count, 0);
 201     s->out_convert= swr_audio_convert_alloc(s->out_sample_fmt,
 202                                             s->int_sample_fmt, s->out.ch_count, 0);
 203
 204
 205     s->postin= s->in;
 206     s->preout= s->out;
 207     s->midbuf= s->in;
 208     s->in_buffer= s->in;
 209     if(!s->resample_first){
 210         s->midbuf.ch_count= s->out.ch_count;
 211         s->in_buffer.ch_count = s->out.ch_count;
 212     }
 213
 214     s->in_buffer.bps = s->postin.bps = s->midbuf.bps = s->preout.bps =  s->int_bps;
 215     s->in_buffer.planar = s->postin.planar = s->midbuf.planar = s->preout.planar =  1;
 216
 217
 218     if(s->rematrix && swr_rematrix_init(s)<0)
 219         return -1;
 220
 221     return 0;
 222 }
 223
 224 static int realloc_audio(AudioData *a, int count){
 225     int i, countb;
 226     AudioData old;
 227
 228     if(a->count >= count)
 229         return 0;
 230
 231     count*=2;
 232
 233     countb= FFALIGN(count*a->bps, 32);
 234     old= *a;
 235
 236     av_assert0(a->planar);
 237     av_assert0(a->bps);
 238     av_assert0(a->ch_count);
 239
 240     a->data= av_malloc(countb*a->ch_count);
 241     if(!a->data)
 242         return AVERROR(ENOMEM);
 243     for(i=0; i<a->ch_count; i++){
 244         a->ch[i]= a->data + i*(a->planar ? countb : a->bps);
 245         if(a->planar) memcpy(a->ch[i], old.ch[i], a->count*a->bps);
 246     }
 247     av_free(old.data);
 248     a->count= count;
 249
 250     return 1;
 251 }
 252
 253 static void copy(AudioData *out, AudioData *in,
 254                  int count){
 255     av_assert0(out->planar == in->planar);
 256     av_assert0(out->bps == in->bps);
 257     av_assert0(out->ch_count == in->ch_count);
 258     if(out->planar){
 259         int ch;
 260         for(ch=0; ch<out->ch_count; ch++)
 261             memcpy(out->ch[ch], in->ch[ch], count*out->bps);
 262     }else
 263         memcpy(out->ch[0], in->ch[0], count*out->ch_count*out->bps);
 264 }
 265
 266 static void fill_audiodata(AudioData *out, uint8_t *in_arg [SWR_CH_MAX]){
 267     int i;
 268     if(out->planar){
 269         for(i=0; i<out->ch_count; i++)
 270             out->ch[i]= in_arg[i];
 271     }else{
 272         for(i=0; i<out->ch_count; i++)
 273             out->ch[i]= in_arg[0] + i*out->bps;
 274     }
 275 }
 276
 277 int swr_convert(struct SwrContext *s, uint8_t *out_arg[SWR_CH_MAX], int out_count,
 278                          const uint8_t *in_arg [SWR_CH_MAX], int  in_count){
 279     AudioData *postin, *midbuf, *preout;
 280     int ret, i/*, in_max*/;
 281     AudioData * in= &s->in;
 282     AudioData *out= &s->out;
 283     AudioData preout_tmp, midbuf_tmp;
 284
 285     if(!s->resample){
 286         if(in_count > out_count)
 287             return -1;
 288         out_count = in_count;
 289     }
 290
 291     fill_audiodata(in ,  in_arg);
 292     fill_audiodata(out, out_arg);
 293
 294 //     in_max= out_count*(int64_t)s->in_sample_rate / s->out_sample_rate + resample_filter_taps;
 295 //     in_count= FFMIN(in_count, in_in + 2 - s->hist_buffer_count);
 296
 297     if((ret=realloc_audio(&s->postin, in_count))<0)
 298         return ret;
 299     if(s->resample_first){
 300         av_assert0(s->midbuf.ch_count ==  s-> in.ch_count);
 301         if((ret=realloc_audio(&s->midbuf, out_count))<0)
 302             return ret;
 303     }else{
 304         av_assert0(s->midbuf.ch_count ==  s->out.ch_count);
 305         if((ret=realloc_audio(&s->midbuf,  in_count))<0)
 306             return ret;
 307     }
 308     if((ret=realloc_audio(&s->preout, out_count))<0)
 309         return ret;
 310
 311     postin= &s->postin;
 312
 313     midbuf_tmp= s->midbuf;
 314     midbuf= &midbuf_tmp;
 315     preout_tmp= s->preout;
 316     preout= &preout_tmp;
 317
 318     if(s->int_sample_fmt == s-> in_sample_fmt && s->in.planar)
 319         postin= in;
 320
 321     if(s->resample_first ? !s->resample : !s->rematrix)
 322         midbuf= postin;
 323
 324     if(s->resample_first ? !s->rematrix : !s->resample)
 325         preout= midbuf;
 326
 327     if(s->int_sample_fmt == s->out_sample_fmt && s->out.planar){
 328         if(preout==in){
 329             out_count= FFMIN(out_count, in_count); //TODO check at teh end if this is needed or redundant
 330             av_assert0(s->in.planar); //we only support planar internally so it has to be, we support copying non planar though
 331             copy(out, in, out_count);
 332             return out_count;
 333         }
 334         else if(preout==postin) preout= midbuf= postin= out;
 335         else if(preout==midbuf) preout= midbuf= out;
 336         else                    preout= out;
 337     }
 338
 339     if(in != postin){
 340         swr_audio_convert(s->in_convert, postin, in, in_count);
 341     }
 342
 343     if(s->resample_first){
 344         if(postin != midbuf)
 345             out_count= resample(s, midbuf, out_count, postin, in_count);
 346         if(midbuf != preout)
 347             swr_rematrix(s, preout, midbuf, out_count, preout==out);
 348     }else{
 349         if(postin != midbuf)
 350             swr_rematrix(s, midbuf, postin, in_count, midbuf==out);
 351         if(midbuf != preout)
 352             out_count= resample(s, preout, out_count, midbuf, in_count);
 353     }
 354
 355     if(preout != out){
 356 //FIXME packed doesnt need more than 1 chan here!
 357         swr_audio_convert(s->out_convert, out, preout, out_count);
 358     }
 359     return out_count;
 360 }
 361
 362 /**
 363  *
 364  * out may be equal in.
 365  */
 366 static void buf_set(AudioData *out, AudioData *in, int count){
 367     if(in->planar){
 368         int ch;
 369         for(ch=0; ch<out->ch_count; ch++)
 370             out->ch[ch]= in->ch[ch] + count*out->bps;
 371     }else
 372         out->ch[0]= in->ch[0] + count*out->ch_count*out->bps;
 373 }
 374
 375 /**
 376  *
 377  * @return number of samples output per channel
 378  */
 379 static int resample(SwrContext *s, AudioData *out_param, int out_count,
 380                              const AudioData * in_param, int in_count){
 381     AudioData in, out, tmp;
 382     int ret_sum=0;
 383     int border=0;
 384
 385     tmp=out=*out_param;
 386     in =  *in_param;
 387
 388     do{
 389         int ret, size, consumed;
 390         if(!s->resample_in_constraint && s->in_buffer_count){
 391             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 392             ret= swr_multiple_resample(s->resample, &out, out_count, &tmp, s->in_buffer_count, &consumed);
 393             out_count -= ret;
 394             ret_sum += ret;
 395             buf_set(&out, &out, ret);
 396             s->in_buffer_count -= consumed;
 397             s->in_buffer_index += consumed;
 398
 399             if(!in_count)
 400                 break;
 401             if(s->in_buffer_count <= border){
 402                 buf_set(&in, &in, -s->in_buffer_count);
 403                 in_count += s->in_buffer_count;
 404                 s->in_buffer_count=0;
 405                 s->in_buffer_index=0;
 406                 border = 0;
 407             }
 408         }
 409
 410         if(in_count && !s->in_buffer_count){
 411             s->in_buffer_index=0;
 412             ret= swr_multiple_resample(s->resample, &out, out_count, &in, in_count, &consumed);
 413             out_count -= ret;
 414             ret_sum += ret;
 415             buf_set(&out, &out, ret);
 416             in_count -= consumed;
 417             buf_set(&in, &in, consumed);
 418         }
 419
 420         //TODO is this check sane considering the advanced copy avoidance below
 421         size= s->in_buffer_index + s->in_buffer_count + in_count;
 422         if(   size > s->in_buffer.count
 423            && s->in_buffer_count + in_count <= s->in_buffer_index){
 424             buf_set(&tmp, &s->in_buffer, s->in_buffer_index);
 425             copy(&s->in_buffer, &tmp, s->in_buffer_count);
 426             s->in_buffer_index=0;
 427         }else
 428             if((ret=realloc_audio(&s->in_buffer, size)) < 0)
 429                 return ret;
 430
 431         if(in_count){
 432             int count= in_count;
 433             if(s->in_buffer_count && s->in_buffer_count+2 < count && out_count) count= s->in_buffer_count+2;
 434
 435             buf_set(&tmp, &s->in_buffer, s->in_buffer_index + s->in_buffer_count);
 436             copy(&tmp, &in, /*in_*/count);
 437             s->in_buffer_count += count;
 438             in_count -= count;
 439             border += count;
 440             buf_set(&in, &in, count);
 441             s->resample_in_constraint= 0;
 442             if(s->in_buffer_count != count || in_count)
 443                 continue;
 444         }
 445         break;
 446     }while(1);
 447
 448     s->resample_in_constraint= !!out_count;
 449
 450     return ret_sum;
 451 }