generic butterfly slightly slower -- hmmm

This commit is contained in:
Mark Borgerding 2003-11-07 02:39:49 +00:00
parent a296b09dbf
commit e9095a161c
3 changed files with 38 additions and 23 deletions

View File

@ -287,31 +287,44 @@ void bfly_generic(
int p
)
{
int u,k,q1,q;
int u,q,d;
kiss_fft_cpx * scratch = st->scratch;
kiss_fft_cpx * twiddles = st->twiddles;
kiss_fft_cpx t;
int Norig = st->nfft;
kiss_fft_cpx * tw = st->twiddles;
kiss_fft_cpx tlo,t1,t2,t3,t4;
for ( u=0; u<m; ++u ) {
k=u;
for ( q1=0 ; q1<p ; ++q1 ) {
scratch[q1] = Fout[ k ];
C_FIXDIV(scratch[q1],p);
k += m;
scratch[0] = Fout[u];
for ( q=1 ; q<p ; ++q ) {
C_MUL( scratch[q] , Fout[ m*q + u] , tw[u*fstride*q] );
Fout[ m*q + u] = scratch[0];
C_ADDTO(Fout[u] , scratch[q] );
}
k=u;
for ( q1=0 ; q1<p ; ++q1 ) {
int twidx=0;
Fout[ k ] = scratch[0];
for (q=1;q<p;++q ) {
twidx += fstride * k;
if (twidx>=Norig) twidx-=Norig;
C_MUL(t,scratch[q] , twiddles[twidx] );
C_ADDTO( Fout[ k ] ,t);
for ( q=1; q<p ; ++q ) {
int twidx = q*fstride * m;
for ( d=1; d<=p/2;++d) {
tlo = tw[ twidx];
twidx += d*fstride*m;
if (twidx >= st->nfft)
twidx -= st->nfft;
t3.r = scratch[q].r * tlo.r;
t3.i = scratch[q].r * tlo.i;
t4.r = scratch[q].i * tlo.i;
t4.i = - scratch[q].i * tlo.r;
t1.r = t3.r - t4.r;
t1.i = t3.i - t4.i;
t2.r = t3.r + t4.r;
t2.i = - (t3.i + t4.i);
C_ADDTO( Fout[ m*d+u ] ,t1);
C_ADDTO( Fout[ m*(p-d)+u ] ,t2);
}
k += m;
}
}
}
@ -338,7 +351,9 @@ void fft_work(
switch (p) {
case 2: bfly2(Fout,fstride,st,m); break;
#if 0
case 3: bfly3(Fout,fstride,st,m); break;
#endif
case 4: bfly4(Fout,fstride,st,m); break;
case 5: bfly5(Fout,fstride,st,m); break;
default: bfly_generic(Fout,fstride,st,m,p); break;

View File

@ -42,9 +42,9 @@ POW3=243 729 2187
POW5=25 125 625 3125
mtime: all bm_fftw
@for n in $(NFFT) $(POW2) $(POW3) $(POW5) ;do \
for n in $(POW3) ;do \
./$(BENCH) -x $(NUMFFTS) -n $$n;\
[ -x ./bm_fftw ] && ./bm_fftw -x $(NUMFFTS) -n $$n ; \
[ "$(DATATYPE)" == "double" ] && [ -x ./bm_fftw ] && ./bm_fftw -x $(NUMFFTS) -n $$n || true ; \
done
snr: all

View File

@ -42,9 +42,9 @@ POW3=243 729 2187
POW5=25 125 625 3125
mtime: all bm_fftw
@for n in $(NFFT) $(POW2) $(POW3) $(POW5) ;do \
for n in $(POW3) ;do \
./$(BENCH) -x $(NUMFFTS) -n $$n;\
[ -x ./bm_fftw ] && ./bm_fftw -x $(NUMFFTS) -n $$n ; \
[ "$(DATATYPE)" == "double" ] && [ -x ./bm_fftw ] && ./bm_fftw -x $(NUMFFTS) -n $$n || true ; \
done
snr: all