sprod3du,dprod3du(3F) sprod3du,dprod3du(3F)NAME
sprod3du, dprod3du - Compute the product of a 3D Fourier transforms with
a 3D filter.
SYNOPSYS
Fortran :
subroutine sprod3du( n1,n2,n3,array,la1,la2,filter,lf1,lf2)
integer n1,n2,n3,la1,la2,lf1,lf2
real array(la1,la2,n3), filter(lf1,lf2,n3)
subroutine dprod3du( n1,n2,n3,array,la1,la2,filter,lf1,lf2)
integer n1,n2,n3,la1,la2,lf1,lf2
real*8 array(la1,la2,n3), filter(lf1,lf2,n3)
C :
#include <fft.h>
int sprod3du(int n1,int n2,int n3,float *array,int la1, int la2,
float *filter, int lf1, int lf2);
int dprod3du(int n1,int n2,int n3,double *array,int la1, int la2,
double *filter, int lf1, int lf2);
DESCRIPTION
sprod3du and dprod3du compute the product of the Fourier transforms of 3D
real sequence (size N1xN2xN3) with the Fourier transform of 3D real
filter. Note, the product of the Fourier transforms of two sequences is
equal to the Fourier transform of their convolution.
PARAMETERS
N1 Integer, the first dimension size of the 3D sequence.
Unchanged on exit.
N2 Integer, the second dimension size of the 3D sequence. Unchanged
on exit.
N3 Integer, the thrid dimension size of the 3D sequence. Unchanged
on exit. ARRAY Array containing the samples of the 3D sequence to be
transformed.
On input, the element {i,j,k} of the sequence is stored as A(i,j,k) in
Fortran , and A[i+j*la1+k*la1*la2] in C.
On exit, the array is overwritten.
LA1 Integer, first leading dimension: increment between the samples of
two consecutive 1D sub-sequences (e.g between {i,j+1,k} and {i,j,k} ).
Unchanged on exit.
LA2 Integer, second leading dimension: number of the 1D sub-sequence
between two consecutive 2D sub-sequences (e.g between {i,j,k+1} and
{i,j,k}). Unchanged on exit.
Page 1
sprod3du,dprod3du(3F) sprod3du,dprod3du(3F)
FILTER Array containing the Fourier Transform of the 3D filter.
Unchanged on exit.
LF1 Integer, filter array first leading dimension. Unchanged on exit.
LF2 Integer, filter array second leading dimension. Unchanged on exit.
Example of Calling Sequence
Given a real sequence of size 100x64x125. We apply successvely the
Direct FFT, the Inverse FFT, and scale back. The elements of each
sequence are stored with increment (stride) 1, the offset between the
first element of two succesive 1D sub-sequences (first leading dimension)
is 102, and the number of 1D sub-sequence between two succesive 2D sub-
sequences (second leading dimension) is 64.
Note : 102 >= 100+2 , and 64 >= 64.
Fortran
real array(0:102-1,0:64-1,0:125-1)
real filter(0:102-1,0:64-1,0:125-1)
real coeff(102+15 + 2*(64+15) + 2*(125+15))
call scfft3dui( 100, 64, 125, coeff)
call scfft3du( -1, 100, 64, 125, array, 102, 64, coeff)
call sprod3du( 100,64,125,array,102,64,filter,102,64)
call csfft3du( 1, 100, 64, 125, array, 102, 64, coeff)
C
#include <fft.h>
float array[64*102*125], filter[64*102*125], *coeff;
coeff = scfft3dui( 100, 64, 125, NULL);
scfft3du( -1, 100, 64, 125, array, 102, 64, coeff);
sprod3du( 100,64,125,array,102,64,filter,102,64);
csfft3du( 1, 100, 64, 125, array, 102, 64, coeff);
NOTE : As the FFTs are not normalized, a successive direct, then inverse
transform, scales the original input by the size of the sequence. Rather
than calling sscal3d or dscal3d to scale back the result, this scaling
factor could be directly applied to the filter transform, thus saving
some extra work.
SEE ALSO
fft, scfft3dui, dzfft3dui, scfft3du, dzfft3du, sscal3d, dscal3d
Page 2
