mirror/kissfft
1
0
Fork 0
mirror of https://github.com/mborgerding/kissfft.git synced 2025-06-04 01:28:23 -04:00
Code Issues Packages Projects Releases Wiki Activity

slightly faster

Browse Source
This commit is contained in:
Mark Borgerding 2003-11-07 03:31:59 +00:00
parent 3a5791f203
commit ae305ca400
1 changed files with 12 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
kiss_fft.c
View File

@ -287,12 +287,12 @@ void bfly_generic(
int p int p
) )
{ {
int u,q,d,fsm,halfp,mp; int u,q,d,fsm,halfp,mp,uf;
kiss_fft_cpx * scratch = st->scratch; kiss_fft_cpx * scratch = st->scratch;
kiss_fft_cpx * scratch2 = scratch + p; kiss_fft_cpx * scratch2 = scratch + p;
kiss_fft_cpx * tw = st->twiddles; kiss_fft_cpx * tw = st->twiddles;
kiss_fft_cpx tlo,t3,t4; kiss_fft_cpx tlo,t3,t4;
kiss_fft_cpx *Foutlo,*Fouthi; kiss_fft_cpx *Foutlo,*Fouthi,*tw2;
fsm = fstride*m; fsm = fstride*m;
halfp=p/2; halfp=p/2;
@ -313,31 +313,33 @@ void bfly_generic(
for ( u=0; u<m; ++u ) { for ( u=0; u<m; ++u ) {
scratch[0] = Fout[0]; scratch[0] = Fout[0];
scratch2 = scratch + p; scratch2 = scratch + p;
uf=u*fstride;
/* d==0 */
Foutlo=Fout;
for ( q=1 ; q<p ; ++q ) { for ( q=1 ; q<p ; ++q ) {
C_MUL( scratch[q] , Fout[ m*q ] , tw[u*fstride*q] ); Foutlo += m;
Fout[ m*q ] = scratch[0]; C_MUL( scratch[q] , *Foutlo , tw[uf*q] );
*Foutlo = scratch[0];
} }
for ( q=1; q<p ; ++q ) { for ( q=1; q<p ; ++q ) {
C_ADDTO(*Fout , scratch[q] ); C_ADDTO(*Fout , scratch[q] );
Foutlo=Fout; Foutlo=Fout;
Fouthi=Fout + mp; Fouthi=Fout + mp;
for ( d=1; d<=halfp;++d) { for ( d=1; d<=halfp;++d) {
Foutlo += m;
Fouthi -= m;
tlo = *scratch2++; tlo = *scratch2++;
Foutlo += m;
t3.r = scratch[q].r * tlo.r; t3.r = scratch[q].r * tlo.r;
t3.i = scratch[q].r * tlo.i; t3.i = scratch[q].r * tlo.i;
t4.r = scratch[q].i * tlo.i; t4.r = scratch[q].i * tlo.i;
t4.i = - scratch[q].i * tlo.r; t4.i = - scratch[q].i * tlo.r;
Fouthi -= m;
Foutlo->r += t3.r - t4.r; Foutlo->r += t3.r - t4.r;
Fouthi->r += t3.r + t4.r; Fouthi->r += t3.r + t4.r;
Foutlo->i += t3.i - t4.i; Foutlo->i += t3.i - t4.i;