int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda, BLASLONG posX, BLASLONG posY,
#ifdef USE_ALPHA
FLOAT alpha_r, FLOAT alpha_i,
#endif
FLOAT *b){
BLASLONG i, js, offset;
FLOAT data01;
FLOAT *ao1;
lda *= 2;
js = n;
while (js > 0){
offset = posX - posY;
if (offset > 0) ao1 = a + posY * 2 + (posX + 0) * lda; else ao1 = a + (posX + 0) * 2 + posY * lda;
i = m;
while (i > 0) {
if (offset > 0) {
data01 = CMULT(*(ao1 + 0), -*(ao1 + 1));
} else
if (offset < 0) {
data01 = CMULT(*(ao1 + 0), *(ao1 + 1));
} else {
data01 = CMULT(*(ao1 + 0), ZERO);
}
if (offset > 0) ao1 += 2; else ao1 += lda;
b[ 0] = data01;
b ++;
offset --;
i --;
}
posX ++;
js --;
}
return 0;
}
int CNAME(BLASLONG m, BLASLONG n, FLOAT *a, BLASLONG lda,
#ifdef USE_ALPHA
FLOAT alpha_r, FLOAT alpha_i,
#endif
FLOAT *b){
BLASLONG i;
FLOAT *a_offset, a1, a2;
lda *= 2;
while (n > 0) {
a_offset = a;
a += 2;
for (i = 0; i < m; i ++) {
a1 = *(a_offset + 0);
a2 = *(a_offset + 1);
*(b + 0) = CMULT(a1, a2);
a_offset += lda;
b ++;
}
n --;
}
return 0;
}
void fft_shift2( int n, int nup, float af, float * f, float ag, float * g )
{
static int nupold=0 , nuptop=0 ;
static complex * row=NULL , * cf=NULL , * cg=NULL ;
int ii , nby2=nup/2 , n21=nby2+1 ;
complex fac , gac ;
float sf , sg , dk ;
#ifdef RECUR
complex csf , csg ;
#endif
ENTRY("fft_shift2") ;
/* 15 Mar 2001: shift too big ==> return all zeros */
if( (af < -n || af > n) && (ag < -n || ag > n) ){
for( ii=0 ; ii < n ; ii++ ) f[ii] = g[ii] = 0.0 ;
EXRETURN ;
}
/* make new memory for row storage? */
if( nup > nuptop ){
if( row != NULL ){ free(row) ; free(cf) ; free(cg) ; }
row = (complex *) malloc( sizeof(complex) * nup ) ;
cf = (complex *) malloc( sizeof(complex) * n21 ) ;
cg = (complex *) malloc( sizeof(complex) * n21 ) ;
nuptop = nup ;
}
/* FFT the pair of rows */
if( g != NULL )
for( ii=0 ; ii < n ; ii++ ){ row[ii].r = f[ii] ; row[ii].i = g[ii] ; }
else
for( ii=0 ; ii < n ; ii++ ){ row[ii].r = f[ii] ; row[ii].i = 0 ; }
#ifdef ZFILL
for( ii=n ; ii < nup ; ii++ ){ row[ii].r = row[ii].i = 0.0 ; }
#else
if( nup > n ){
sf = 0.5 * (row[0].r + row[n-1].r) ; sg = 0.5 * (row[0].i + row[n-1].i) ;
for( ii=n ; ii < nup ; ii++ ){ row[ii].r = sf ; row[ii].i = sg ; }
}
#endif
csfft_cox( -1 , nup , row ) ;
/* untangle FFT coefficients from row into cf,cg */
cf[0].r = 2.0 * row[0].r ; cf[0].i = 0.0 ; /* twice too big */
cg[0].r = 2.0 * row[0].i ; cg[0].i = 0.0 ;
for( ii=1 ; ii < nby2 ; ii++ ){
cf[ii].r = row[ii].r + row[nup-ii].r ;
cf[ii].i = row[ii].i - row[nup-ii].i ;
cg[ii].r = row[ii].i + row[nup-ii].i ;
cg[ii].i = -row[ii].r + row[nup-ii].r ;
}
cf[nby2].r = 2.0 * row[nby2].r ; cf[nby2].i = 0.0 ;
cg[nby2].r = 2.0 * row[nby2].i ; cg[nby2].i = 0.0 ;
/* phase shift both rows (cf,cg) */
dk = (2.0*PI) / nup ;
sf = -af * dk ; sg = -ag * dk ;
#ifdef RECUR
csf = CEXPIT(sf) ; csg = CEXPIT(sg) ;
fac.r = gac.r = 1.0 ;
fac.i = gac.i = 0.0 ;
#endif
for( ii=1 ; ii <= nby2 ; ii++ ){
#ifdef RECUR
fac = CMULT( csf , fac ) ; cf[ii] = CMULT( fac , cf[ii] ) ;
gac = CMULT( csg , gac ) ; cg[ii] = CMULT( gac , cg[ii] ) ;
#else
fac = CEXPIT(ii*sf) ; cf[ii] = CMULT( fac , cf[ii] ) ;
gac = CEXPIT(ii*sg) ; cg[ii] = CMULT( gac , cg[ii] ) ;
#endif
}
cf[nby2].i = 0.0 ; cg[nby2].i = 0.0 ;
/* retangle the coefficients from 2 rows */
row[0].r = cf[0].r ; row[0].i = cg[0].r ;
for( ii=1 ; ii < nby2 ; ii++ ){
row[ii].r = cf[ii].r - cg[ii].i ;
row[ii].i = cf[ii].i + cg[ii].r ;
row[nup-ii].r = cf[ii].r + cg[ii].i ;
row[nup-ii].i = -cf[ii].i + cg[ii].r ;
}
row[nby2].r = cf[nby2].r ;
row[nby2].i = cg[nby2].r ;
/* inverse FFT and store back in output arrays */
csfft_cox( 1 , nup , row ) ;
sf = 0.5 / nup ; /* 0.5 to allow for twice too big above */
if( g != NULL )
for( ii=0; ii < n; ii++ ){ f[ii] = sf*row[ii].r; g[ii] = sf*row[ii].i; }
else
for( ii=0; ii < n; ii++ ){ f[ii] = sf*row[ii].r; }
EXRETURN ;
}
void ft_shift2( int n, int nup, float af, float * f, float ag, float * g )
{
static int nupold=0 ;
static complex * row=NULL , * cf=NULL , * cg=NULL ;
int ii , nby2=nup/2 , n21=nby2+1 ;
complex fac , gac ;
float sf , sg , dk ;
/* make new memory for row storage? */
if( nup > nupold ){
if( row != NULL ){ free(row) ; free(cf) ; free(cg) ; }
row = (complex *) malloc( sizeof(complex) * nup ) ;
cf = (complex *) malloc( sizeof(complex) * n21 ) ;
cg = (complex *) malloc( sizeof(complex) * n21 ) ;
nupold = nup ;
}
/* FFT the pair of rows */
for( ii=0 ; ii < n ; ii++ ){ row[ii].r = f[ii] ; row[ii].i = g[ii] ; }
for( ; ii < nup ; ii++ ){ row[ii].r = row[ii].i = 0.0 ; }
csfft_cox( -1 , nup , row ) ;
/* untangle FFT coefficients from row into cf,cg */
cf[0].r = 2.0 * row[0].r ; cf[0].i = 0.0 ; /* twice too big */
cg[0].r = 2.0 * row[0].i ; cg[0].i = 0.0 ;
for( ii=1 ; ii < nby2 ; ii++ ){
cf[ii].r = row[ii].r + row[nup-ii].r ;
cf[ii].i = row[ii].i - row[nup-ii].i ;
cg[ii].r = row[ii].i + row[nup-ii].i ;
cg[ii].i = -row[ii].r + row[nup-ii].r ;
}
cf[nby2].r = 2.0 * row[nby2].r ; cf[nby2].i = 0.0 ;
cg[nby2].r = 2.0 * row[nby2].i ; cg[nby2].i = 0.0 ;
/* phase shift both rows (cf,cg) */
dk = (2.0*PI) / nup ;
sf = -af * dk ; sg = -ag * dk ;
for( ii=1 ; ii <= nby2 ; ii++ ){
fac = CEXPIT(ii*sf) ; cf[ii] = CMULT( fac , cf[ii] ) ;
gac = CEXPIT(ii*sg) ; cg[ii] = CMULT( gac , cg[ii] ) ;
}
cf[nby2].i = 0.0 ; cg[nby2].i = 0.0 ;
/* retangle the coefficients from 2 rows */
row[0].r = cf[0].r ; row[0].i = cg[0].r ;
for( ii=1 ; ii < nby2 ; ii++ ){
row[ii].r = cf[ii].r - cg[ii].i ;
row[ii].i = cf[ii].i + cg[ii].r ;
row[nup-ii].r = cf[ii].r + cg[ii].i ;
row[nup-ii].i = -cf[ii].i + cg[ii].r ;
}
row[nby2].r = cf[nby2].r ;
row[nby2].i = cg[nby2].r ;
/* inverse FFT and store back in output arrays */
csfft_cox( 1 , nup , row ) ;
sf = 0.5 / nup ; /* 0.5 to allow for twice too big above */
for( ii=0 ; ii < n ; ii++ ){
f[ii] = sf * row[ii].r ; g[ii] = sf * row[ii].i ;
}
return ;
}
