static void kf_work(
kiss_fft_cpx * Fout,
const kiss_fft_cpx * f,
size_t fstride,
int in_stride,
const celt_int16 * factors,
const kiss_fft_state *st,
int N,
int m2
)
{
const int p=*factors++; /* the radix */
const int m=*factors++; /* stage's fft length/p */
/*printf ("fft %d %d %d %d %d %d %d\n", p*m, m, p, s2, fstride*in_stride, N, m2);*/
if (m!=1)
kf_work( Fout , f, fstride*p, in_stride, factors,st, N*p, m);
/* Compensate for longer twiddles table (when sharing) */
if (st->shift>0)
fstride <<= st->shift;
switch (p) {
case 2: kf_bfly2(Fout,fstride,st,m, N, m2); break;
case 4: kf_bfly4(Fout,fstride,st,m, N, m2); break;
#ifndef RADIX_TWO_ONLY
case 3: kf_bfly3(Fout,fstride,st,m, N, m2); break;
case 5: kf_bfly5(Fout,fstride,st,m, N, m2); break;
#endif
}
}
void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
{
int m2, m;
int p;
int L;
int fstride[MAXFACTORS];
int i;
int shift;
/* st->shift can be -1 */
shift = st->shift>0 ? st->shift : 0;
celt_assert2 (fin != fout, "In-place FFT not supported");
/* Bit-reverse the input */
for (i=0;i<st->nfft;i++)
{
fout[st->bitrev[i]] = fin[i];
#ifndef FIXED_POINT
fout[st->bitrev[i]].r *= st->scale;
fout[st->bitrev[i]].i *= st->scale;
#endif
}
fstride[0] = 1;
L=0;
do {
p = st->factors[2*L];
m = st->factors[2*L+1];
fstride[L+1] = fstride[L]*p;
L++;
} while(m!=1);
m = st->factors[2*L-1];
for (i=L-1;i>=0;i--)
{
if (i!=0)
m2 = st->factors[2*i-1];
else
m2 = 1;
switch (st->factors[2*i])
{
case 2:
kf_bfly2(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
case 4:
kf_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
#ifndef RADIX_TWO_ONLY
case 3:
kf_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
case 5:
kf_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
#endif
}
m = m2;
}
}
static
void kf_work(
kiss_fft_cpx * Fout,
const kiss_fft_cpx * f,
const size_t fstride,
int in_stride,
int * factors,
const kiss_fft_cfg st
)
{
kiss_fft_cpx * Fout_beg=Fout;
const int p=*factors++; /* the radix */
const int m=*factors++; /* stage's fft length/p */
const kiss_fft_cpx * Fout_end = Fout + p*m;
if (m==1) {
do {
*Fout = *f;
f += fstride*in_stride;
} while(++Fout != Fout_end );
} else {
do {
kf_work( Fout , f, fstride*p, in_stride, factors,st);
f += fstride*in_stride;
} while( (Fout += m) != Fout_end );
}
Fout=Fout_beg;
switch (p) {
case 2:
kf_bfly2(Fout,fstride,st,m);
break;
case 3:
kf_bfly3(Fout,fstride,st,m);
break;
case 4:
kf_bfly4(Fout,fstride,st,m);
break;
case 5:
kf_bfly5(Fout,fstride,st,m);
break;
default:
kf_bfly_generic(Fout,fstride,st,m,p);
break;
}
}
void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout)
{
int m2, m;
int p;
int L;
int fstride[MAXFACTORS];
int i;
int shift;
/* st->shift can be -1 */
shift = st->shift>0 ? st->shift : 0;
fstride[0] = 1;
L=0;
do {
p = st->factors[2*L];
m = st->factors[2*L+1];
fstride[L+1] = fstride[L]*p;
L++;
} while(m!=1);
m = st->factors[2*L-1];
for (i=L-1;i>=0;i--)
{
if (i!=0)
m2 = st->factors[2*i-1];
else
m2 = 1;
switch (st->factors[2*i])
{
case 2:
kf_bfly2(fout, m, fstride[i]);
break;
case 4:
kf_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
#ifndef RADIX_TWO_ONLY
case 3:
kf_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
case 5:
kf_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2);
break;
#endif
}
m = m2;
}
}
static
void kf_work(
kiss_fft_cpx * Fout,
const kiss_fft_cpx * f,
const size_t fstride,
int in_stride,
int * factors,
const kiss_fft_cfg st
)
{
kiss_fft_cpx * Fout_beg=Fout;
const int p=*factors++; /* the radix */
const int m=*factors++; /* stage's fft length/p */
const kiss_fft_cpx * Fout_end = Fout + p*m;
if (m==1) {
do{
*Fout = *f;
f += fstride*in_stride;
}while(++Fout != Fout_end );
}else{
do{
// recursive call:
// DFT of size m*p performed by doing
// p instances of smaller DFTs of size m,
// each one takes a decimated version of the input
kf_work( Fout , f, fstride*p, in_stride, factors,st);
f += fstride*in_stride;
}while( (Fout += m) != Fout_end );
}
Fout=Fout_beg;
// recombine the p smaller DFTs
switch (p) {
case 2: kf_bfly2(Fout,fstride,st,m); break;
case 4: kf_bfly4(Fout,fstride,st,m); break;
default: kf_bfly_generic(Fout,fstride,st,m,p); break;
}
}
static
void kf_work(
kiss_fft_cpx * Fout,
const kiss_fft_cpx * f,
const size_t fstride,
int in_stride,
int * factors,
const kiss_fft_cfg st,
int N,
int s2,
int m2
)
{
int i;
kiss_fft_cpx * Fout_beg=Fout;
const int p=*factors++; /* the radix */
const int m=*factors++; /* stage's fft length/p */
#if 0
/*printf ("fft %d %d %d %d %d %d\n", p*m, m, p, s2, fstride*in_stride, N);*/
if (m==1)
{
/* int j;
for (j=0;j<p;j++)
{
Fout[j] = *f;
f += fstride*in_stride;
}*/
} else {
int j;
for (j=0;j<p;j++)
{
kf_work( Fout , f, fstride*p, in_stride, factors,st, N*p, fstride*in_stride, m);
f += fstride*in_stride;
Fout += m;
}
}
Fout=Fout_beg;
switch (p) {
case 2: kf_bfly2(Fout,fstride,st,m); break;
case 3: kf_bfly3(Fout,fstride,st,m); break;
case 4: kf_bfly4(Fout,fstride,st,m); break;
case 5: kf_bfly5(Fout,fstride,st,m); break;
default: kf_bfly_generic(Fout,fstride,st,m,p); break;
}
#else
/*printf ("fft %d %d %d %d %d %d %d\n", p*m, m, p, s2, fstride*in_stride, N, m2);*/
if (m==1)
{
/*for (i=0;i<N;i++)
{
int j;
Fout = Fout_beg+i*m2;
const kiss_fft_cpx * f2 = f+i*s2;
for (j=0;j<p;j++)
{
*Fout++ = *f2;
f2 += fstride*in_stride;
}
}*/
}else{
kf_work( Fout , f, fstride*p, in_stride, factors,st, N*p, fstride*in_stride, m);
}
switch (p) {
case 2: kf_bfly2(Fout,fstride,st,m, N, m2); break;
case 3: for (i=0;i<N;i++){Fout=Fout_beg+i*m2; kf_bfly3(Fout,fstride,st,m);} break;
case 4: kf_bfly4(Fout,fstride,st,m, N, m2); break;
case 5: for (i=0;i<N;i++){Fout=Fout_beg+i*m2; kf_bfly5(Fout,fstride,st,m);} break;
default: for (i=0;i<N;i++){Fout=Fout_beg+i*m2; kf_bfly_generic(Fout,fstride,st,m,p);} break;
}
#endif
}
static
void kf_work(
kiss_fft_cpx * Fout,
const kiss_fft_cpx * f,
const size_t fstride,
int in_stride,
int * factors,
const kiss_fft_cfg st
)
{
kiss_fft_cpx * Fout_beg=Fout;
const int p=*factors++; /* the radix */
const int m=*factors++; /* stage's fft length/p */
const kiss_fft_cpx * Fout_end = Fout + p*m;
#ifdef _OPENMP
// use openmp extensions at the
// top-level (not recursive)
if (fstride==1 && p<=5)
{
int k;
// execute the p different work units in different threads
# pragma omp parallel for
for (k=0;k<p;++k)
kf_work( Fout +k*m, f+ fstride*in_stride*k,fstride*p,in_stride,factors,st);
// all threads have joined by this point
switch (p) {
case 2: kf_bfly2(Fout,fstride,st,m); break;
case 3: kf_bfly3(Fout,fstride,st,m); break;
case 4: kf_bfly4(Fout,fstride,st,m); break;
case 5: kf_bfly5(Fout,fstride,st,m); break;
default: kf_bfly_generic(Fout,fstride,st,m,p); break;
}
return;
}
#endif
if (m==1) {
do{
*Fout = *f;
f += fstride*in_stride;
}while(++Fout != Fout_end );
}else{
do{
// recursive call:
// DFT of size m*p performed by doing
// p instances of smaller DFTs of size m,
// each one takes a decimated version of the input
kf_work( Fout , f, fstride*p, in_stride, factors,st);
f += fstride*in_stride;
}while( (Fout += m) != Fout_end );
}
Fout=Fout_beg;
// recombine the p smaller DFTs
switch (p) {
case 2: kf_bfly2(Fout,fstride,st,m); break;
case 3: kf_bfly3(Fout,fstride,st,m); break;
case 4: kf_bfly4(Fout,fstride,st,m); break;
case 5: kf_bfly5(Fout,fstride,st,m); break;
default: kf_bfly_generic(Fout,fstride,st,m,p); break;
}
}