| 1 |
/* |
| 2 |
* Copyright (c) 2003-2010, Mark Borgerding. All rights reserved. |
| 3 |
* This file is part of KISS FFT - https://github.com/mborgerding/kissfft |
| 4 |
* |
| 5 |
* SPDX-License-Identifier: BSD-3-Clause |
| 6 |
* See COPYING file for more information. |
| 7 |
*/ |
| 8 |
|
| 9 |
|
| 10 |
#include "_kiss_fft_guts.h" |
| 11 |
/* The guts header contains all the multiplication and addition macros that are defined for |
| 12 |
fixed or floating point complex numbers. It also delares the kf_ internal functions. |
| 13 |
*/ |
| 14 |
|
| 15 |
static void kf_bfly2( |
| 16 |
kiss_fft_cpx * Fout, |
| 17 |
const size_t fstride, |
| 18 |
const kiss_fft_cfg st, |
| 19 |
int m |
| 20 |
) |
| 21 |
{ |
| 22 |
kiss_fft_cpx * Fout2; |
| 23 |
kiss_fft_cpx * tw1 = st->twiddles; |
| 24 |
kiss_fft_cpx t; |
| 25 |
Fout2 = Fout + m; |
| 26 |
do{ |
| 27 |
C_FIXDIV(*Fout,2); C_FIXDIV(*Fout2,2); |
| 28 |
|
| 29 |
C_MUL (t, *Fout2 , *tw1); |
| 30 |
tw1 += fstride; |
| 31 |
C_SUB( *Fout2 , *Fout , t ); |
| 32 |
C_ADDTO( *Fout , t ); |
| 33 |
++Fout2; |
| 34 |
++Fout; |
| 35 |
}while (--m); |
| 36 |
} |
| 37 |
|
| 38 |
static void kf_bfly4( |
| 39 |
kiss_fft_cpx * Fout, |
| 40 |
const size_t fstride, |
| 41 |
const kiss_fft_cfg st, |
| 42 |
const size_t m |
| 43 |
) |
| 44 |
{ |
| 45 |
kiss_fft_cpx *tw1,*tw2,*tw3; |
| 46 |
kiss_fft_cpx scratch[6]; |
| 47 |
size_t k=m; |
| 48 |
const size_t m2=2*m; |
| 49 |
const size_t m3=3*m; |
| 50 |
|
| 51 |
|
| 52 |
tw3 = tw2 = tw1 = st->twiddles; |
| 53 |
|
| 54 |
do { |
| 55 |
C_FIXDIV(*Fout,4); C_FIXDIV(Fout[m],4); C_FIXDIV(Fout[m2],4); C_FIXDIV(Fout[m3],4); |
| 56 |
|
| 57 |
C_MUL(scratch[0],Fout[m] , *tw1 ); |
| 58 |
C_MUL(scratch[1],Fout[m2] , *tw2 ); |
| 59 |
C_MUL(scratch[2],Fout[m3] , *tw3 ); |
| 60 |
|
| 61 |
C_SUB( scratch[5] , *Fout, scratch[1] ); |
| 62 |
C_ADDTO(*Fout, scratch[1]); |
| 63 |
C_ADD( scratch[3] , scratch[0] , scratch[2] ); |
| 64 |
C_SUB( scratch[4] , scratch[0] , scratch[2] ); |
| 65 |
C_SUB( Fout[m2], *Fout, scratch[3] ); |
| 66 |
tw1 += fstride; |
| 67 |
tw2 += fstride*2; |
| 68 |
tw3 += fstride*3; |
| 69 |
C_ADDTO( *Fout , scratch[3] ); |
| 70 |
|
| 71 |
if(st->inverse) { |
| 72 |
Fout[m].r = scratch[5].r - scratch[4].i; |
| 73 |
Fout[m].i = scratch[5].i + scratch[4].r; |
| 74 |
Fout[m3].r = scratch[5].r + scratch[4].i; |
| 75 |
Fout[m3].i = scratch[5].i - scratch[4].r; |
| 76 |
}else{ |
| 77 |
Fout[m].r = scratch[5].r + scratch[4].i; |
| 78 |
Fout[m].i = scratch[5].i - scratch[4].r; |
| 79 |
Fout[m3].r = scratch[5].r - scratch[4].i; |
| 80 |
Fout[m3].i = scratch[5].i + scratch[4].r; |
| 81 |
} |
| 82 |
++Fout; |
| 83 |
}while(--k); |
| 84 |
} |
| 85 |
|
| 86 |
static void kf_bfly3( |
| 87 |
kiss_fft_cpx * Fout, |
| 88 |
const size_t fstride, |
| 89 |
const kiss_fft_cfg st, |
| 90 |
size_t m |
| 91 |
) |
| 92 |
{ |
| 93 |
size_t k=m; |
| 94 |
const size_t m2 = 2*m; |
| 95 |
kiss_fft_cpx *tw1,*tw2; |
| 96 |
kiss_fft_cpx scratch[5]; |
| 97 |
kiss_fft_cpx epi3; |
| 98 |
epi3 = st->twiddles[fstride*m]; |
| 99 |
|
| 100 |
tw1=tw2=st->twiddles; |
| 101 |
|
| 102 |
do{ |
| 103 |
C_FIXDIV(*Fout,3); C_FIXDIV(Fout[m],3); C_FIXDIV(Fout[m2],3); |
| 104 |
|
| 105 |
C_MUL(scratch[1],Fout[m] , *tw1); |
| 106 |
C_MUL(scratch[2],Fout[m2] , *tw2); |
| 107 |
|
| 108 |
C_ADD(scratch[3],scratch[1],scratch[2]); |
| 109 |
C_SUB(scratch[0],scratch[1],scratch[2]); |
| 110 |
tw1 += fstride; |
| 111 |
tw2 += fstride*2; |
| 112 |
|
| 113 |
Fout[m].r = Fout->r - HALF_OF(scratch[3].r); |
| 114 |
Fout[m].i = Fout->i - HALF_OF(scratch[3].i); |
| 115 |
|
| 116 |
C_MULBYSCALAR( scratch[0] , epi3.i ); |
| 117 |
|
| 118 |
C_ADDTO(*Fout,scratch[3]); |
| 119 |
|
| 120 |
Fout[m2].r = Fout[m].r + scratch[0].i; |
| 121 |
Fout[m2].i = Fout[m].i - scratch[0].r; |
| 122 |
|
| 123 |
Fout[m].r -= scratch[0].i; |
| 124 |
Fout[m].i += scratch[0].r; |
| 125 |
|
| 126 |
++Fout; |
| 127 |
}while(--k); |
| 128 |
} |
| 129 |
|
| 130 |
static void kf_bfly5( |
| 131 |
kiss_fft_cpx * Fout, |
| 132 |
const size_t fstride, |
| 133 |
const kiss_fft_cfg st, |
| 134 |
int m |
| 135 |
) |
| 136 |
{ |
| 137 |
kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; |
| 138 |
int u; |
| 139 |
kiss_fft_cpx scratch[13]; |
| 140 |
kiss_fft_cpx * twiddles = st->twiddles; |
| 141 |
kiss_fft_cpx *tw; |
| 142 |
kiss_fft_cpx ya,yb; |
| 143 |
ya = twiddles[fstride*m]; |
| 144 |
yb = twiddles[fstride*2*m]; |
| 145 |
|
| 146 |
Fout0=Fout; |
| 147 |
Fout1=Fout0+m; |
| 148 |
Fout2=Fout0+2*m; |
| 149 |
Fout3=Fout0+3*m; |
| 150 |
Fout4=Fout0+4*m; |
| 151 |
|
| 152 |
tw=st->twiddles; |
| 153 |
for ( u=0; u<m; ++u ) { |
| 154 |
C_FIXDIV( *Fout0,5); C_FIXDIV( *Fout1,5); C_FIXDIV( *Fout2,5); C_FIXDIV( *Fout3,5); C_FIXDIV( *Fout4,5); |
| 155 |
scratch[0] = *Fout0; |
| 156 |
|
| 157 |
C_MUL(scratch[1] ,*Fout1, tw[u*fstride]); |
| 158 |
C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]); |
| 159 |
C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]); |
| 160 |
C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]); |
| 161 |
|
| 162 |
C_ADD( scratch[7],scratch[1],scratch[4]); |
| 163 |
C_SUB( scratch[10],scratch[1],scratch[4]); |
| 164 |
C_ADD( scratch[8],scratch[2],scratch[3]); |
| 165 |
C_SUB( scratch[9],scratch[2],scratch[3]); |
| 166 |
|
| 167 |
Fout0->r += scratch[7].r + scratch[8].r; |
| 168 |
Fout0->i += scratch[7].i + scratch[8].i; |
| 169 |
|
| 170 |
scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r); |
| 171 |
scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r); |
| 172 |
|
| 173 |
scratch[6].r = S_MUL(scratch[10].i,ya.i) + S_MUL(scratch[9].i,yb.i); |
| 174 |
scratch[6].i = -S_MUL(scratch[10].r,ya.i) - S_MUL(scratch[9].r,yb.i); |
| 175 |
|
| 176 |
C_SUB(*Fout1,scratch[5],scratch[6]); |
| 177 |
C_ADD(*Fout4,scratch[5],scratch[6]); |
| 178 |
|
| 179 |
scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r); |
| 180 |
scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r); |
| 181 |
scratch[12].r = - S_MUL(scratch[10].i,yb.i) + S_MUL(scratch[9].i,ya.i); |
| 182 |
scratch[12].i = S_MUL(scratch[10].r,yb.i) - S_MUL(scratch[9].r,ya.i); |
| 183 |
|
| 184 |
C_ADD(*Fout2,scratch[11],scratch[12]); |
| 185 |
C_SUB(*Fout3,scratch[11],scratch[12]); |
| 186 |
|
| 187 |
++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; |
| 188 |
} |
| 189 |
} |
| 190 |
|
| 191 |
/* perform the butterfly for one stage of a mixed radix FFT */ |
| 192 |
static void kf_bfly_generic( |
| 193 |
kiss_fft_cpx * Fout, |
| 194 |
const size_t fstride, |
| 195 |
const kiss_fft_cfg st, |
| 196 |
int m, |
| 197 |
int p |
| 198 |
) |
| 199 |
{ |
| 200 |
int u,k,q1,q; |
| 201 |
kiss_fft_cpx * twiddles = st->twiddles; |
| 202 |
kiss_fft_cpx t; |
| 203 |
int Norig = st->nfft; |
| 204 |
|
| 205 |
kiss_fft_cpx * scratch = (kiss_fft_cpx*)KISS_FFT_TMP_ALLOC(sizeof(kiss_fft_cpx)*p); |
| 206 |
|
| 207 |
for ( u=0; u<m; ++u ) { |
| 208 |
k=u; |
| 209 |
for ( q1=0 ; q1<p ; ++q1 ) { |
| 210 |
scratch[q1] = Fout[ k ]; |
| 211 |
C_FIXDIV(scratch[q1],p); |
| 212 |
k += m; |
| 213 |
} |
| 214 |
|
| 215 |
k=u; |
| 216 |
for ( q1=0 ; q1<p ; ++q1 ) { |
| 217 |
int twidx=0; |
| 218 |
Fout[ k ] = scratch[0]; |
| 219 |
for (q=1;q<p;++q ) { |
| 220 |
twidx += fstride * k; |
| 221 |
if (twidx>=Norig) twidx-=Norig; |
| 222 |
C_MUL(t,scratch[q] , twiddles[twidx] ); |
| 223 |
C_ADDTO( Fout[ k ] ,t); |
| 224 |
} |
| 225 |
k += m; |
| 226 |
} |
| 227 |
} |
| 228 |
KISS_FFT_TMP_FREE(scratch); |
| 229 |
} |
| 230 |
|
| 231 |
static |
| 232 |
void kf_work( |
| 233 |
kiss_fft_cpx * Fout, |
| 234 |
const kiss_fft_cpx * f, |
| 235 |
const size_t fstride, |
| 236 |
int in_stride, |
| 237 |
int * factors, |
| 238 |
const kiss_fft_cfg st |
| 239 |
) |
| 240 |
{ |
| 241 |
kiss_fft_cpx * Fout_beg=Fout; |
| 242 |
const int p=*factors++; /* the radix */ |
| 243 |
const int m=*factors++; /* stage's fft length/p */ |
| 244 |
const kiss_fft_cpx * Fout_end = Fout + p*m; |
| 245 |
|
| 246 |
#ifdef _OPENMP |
| 247 |
// use openmp extensions at the |
| 248 |
// top-level (not recursive) |
| 249 |
if (fstride==1 && p<=5) |
| 250 |
{ |
| 251 |
int k; |
| 252 |
|
| 253 |
// execute the p different work units in different threads |
| 254 |
# pragma omp parallel for |
| 255 |
for (k=0;k<p;++k) |
| 256 |
kf_work( Fout +k*m, f+ fstride*in_stride*k,fstride*p,in_stride,factors,st); |
| 257 |
// all threads have joined by this point |
| 258 |
|
| 259 |
switch (p) { |
| 260 |
case 2: kf_bfly2(Fout,fstride,st,m); break; |
| 261 |
case 3: kf_bfly3(Fout,fstride,st,m); break; |
| 262 |
case 4: kf_bfly4(Fout,fstride,st,m); break; |
| 263 |
case 5: kf_bfly5(Fout,fstride,st,m); break; |
| 264 |
default: kf_bfly_generic(Fout,fstride,st,m,p); break; |
| 265 |
} |
| 266 |
return; |
| 267 |
} |
| 268 |
#endif |
| 269 |
|
| 270 |
if (m==1) { |
| 271 |
do{ |
| 272 |
*Fout = *f; |
| 273 |
f += fstride*in_stride; |
| 274 |
}while(++Fout != Fout_end ); |
| 275 |
}else{ |
| 276 |
do{ |
| 277 |
// recursive call: |
| 278 |
// DFT of size m*p performed by doing |
| 279 |
// p instances of smaller DFTs of size m, |
| 280 |
// each one takes a decimated version of the input |
| 281 |
kf_work( Fout , f, fstride*p, in_stride, factors,st); |
| 282 |
f += fstride*in_stride; |
| 283 |
}while( (Fout += m) != Fout_end ); |
| 284 |
} |
| 285 |
|
| 286 |
Fout=Fout_beg; |
| 287 |
|
| 288 |
// recombine the p smaller DFTs |
| 289 |
switch (p) { |
| 290 |
case 2: kf_bfly2(Fout,fstride,st,m); break; |
| 291 |
case 3: kf_bfly3(Fout,fstride,st,m); break; |
| 292 |
case 4: kf_bfly4(Fout,fstride,st,m); break; |
| 293 |
case 5: kf_bfly5(Fout,fstride,st,m); break; |
| 294 |
default: kf_bfly_generic(Fout,fstride,st,m,p); break; |
| 295 |
} |
| 296 |
} |
| 297 |
|
| 298 |
/* facbuf is populated by p1,m1,p2,m2, ... |
| 299 |
where |
| 300 |
p[i] * m[i] = m[i-1] |
| 301 |
m0 = n */ |
| 302 |
static |
| 303 |
void kf_factor(int n,int * facbuf) |
| 304 |
{ |
| 305 |
int p=4; |
| 306 |
double floor_sqrt; |
| 307 |
floor_sqrt = floor( sqrt((double)n) ); |
| 308 |
|
| 309 |
/*factor out powers of 4, powers of 2, then any remaining primes */ |
| 310 |
do { |
| 311 |
while (n % p) { |
| 312 |
switch (p) { |
| 313 |
case 4: p = 2; break; |
| 314 |
case 2: p = 3; break; |
| 315 |
default: p += 2; break; |
| 316 |
} |
| 317 |
if (p > floor_sqrt) |
| 318 |
p = n; /* no more factors, skip to end */ |
| 319 |
} |
| 320 |
n /= p; |
| 321 |
*facbuf++ = p; |
| 322 |
*facbuf++ = n; |
| 323 |
} while (n > 1); |
| 324 |
} |
| 325 |
|
| 326 |
/* |
| 327 |
* |
| 328 |
* User-callable function to allocate all necessary storage space for the fft. |
| 329 |
* |
| 330 |
* The return value is a contiguous block of memory, allocated with malloc. As such, |
| 331 |
* It can be freed with free(), rather than a kiss_fft-specific function. |
| 332 |
* */ |
| 333 |
kiss_fft_cfg kiss_fft_alloc(int nfft,int inverse_fft,void * mem,size_t * lenmem ) |
| 334 |
{ |
| 335 |
kiss_fft_cfg st=NULL; |
| 336 |
size_t memneeded = sizeof(struct kiss_fft_state) |
| 337 |
+ sizeof(kiss_fft_cpx)*(nfft-1); /* twiddle factors*/ |
| 338 |
|
| 339 |
if ( lenmem==NULL ) { |
| 340 |
st = ( kiss_fft_cfg)KISS_FFT_MALLOC( memneeded ); |
| 341 |
}else{ |
| 342 |
if (mem != NULL && *lenmem >= memneeded) |
| 343 |
st = (kiss_fft_cfg)mem; |
| 344 |
*lenmem = memneeded; |
| 345 |
} |
| 346 |
if (st) { |
| 347 |
int i; |
| 348 |
st->nfft=nfft; |
| 349 |
st->inverse = inverse_fft; |
| 350 |
|
| 351 |
for (i=0;i<nfft;++i) { |
| 352 |
const double pi=3.141592653589793238462643383279502884197169399375105820974944; |
| 353 |
double phase = -2*pi*i / nfft; |
| 354 |
if (st->inverse) |
| 355 |
phase *= -1; |
| 356 |
kf_cexp(st->twiddles+i, phase ); |
| 357 |
} |
| 358 |
|
| 359 |
kf_factor(nfft,st->factors); |
| 360 |
} |
| 361 |
return st; |
| 362 |
} |
| 363 |
|
| 364 |
|
| 365 |
void kiss_fft_stride(kiss_fft_cfg st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout,int in_stride) |
| 366 |
{ |
| 367 |
if (fin == fout) { |
| 368 |
//NOTE: this is not really an in-place FFT algorithm. |
| 369 |
//It just performs an out-of-place FFT into a temp buffer |
| 370 |
kiss_fft_cpx * tmpbuf = (kiss_fft_cpx*)KISS_FFT_TMP_ALLOC( sizeof(kiss_fft_cpx)*st->nfft); |
| 371 |
kf_work(tmpbuf,fin,1,in_stride, st->factors,st); |
| 372 |
memcpy(fout,tmpbuf,sizeof(kiss_fft_cpx)*st->nfft); |
| 373 |
KISS_FFT_TMP_FREE(tmpbuf); |
| 374 |
}else{ |
| 375 |
kf_work( fout, fin, 1,in_stride, st->factors,st ); |
| 376 |
} |
| 377 |
} |
| 378 |
|
| 379 |
void kiss_fft(kiss_fft_cfg cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) |
| 380 |
{ |
| 381 |
kiss_fft_stride(cfg,fin,fout,1); |
| 382 |
} |
| 383 |
|
| 384 |
|
| 385 |
void kiss_fft_cleanup(void) |
| 386 |
{ |
| 387 |
// nothing needed any more |
| 388 |
} |
| 389 |
|
| 390 |
int kiss_fft_next_fast_size(int n) |
| 391 |
{ |
| 392 |
while(1) { |
| 393 |
int m=n; |
| 394 |
while ( (m%2) == 0 ) m/=2; |
| 395 |
while ( (m%3) == 0 ) m/=3; |
| 396 |
while ( (m%5) == 0 ) m/=5; |
| 397 |
if (m<=1) |
| 398 |
break; /* n is completely factorable by twos, threes, and fives */ |
| 399 |
n++; |
| 400 |
} |
| 401 |
return n; |
| 402 |
} |