cffts2(3P) Sun Performance Library cffts2(3P)NAMEcffts2 - initialize the trigonometric weight and factor tables or com‐
pute the two-dimensional inverse Fast Fourier Transform of a two-dimen‐
sional complex array.
SYNOPSIS
SUBROUTINE CFFTS2(IOPT, N1, N2, SCALE, X, LDX, Y, LDY, TRIGS, IFAC, WORK, LWORK, IERR)
INTEGER IOPT, N1, N2, LDX, LDY, IFAC(*), LWORK, IERR
COMPLEX X(LDX, *)
REAL SCALE, Y(LDY, *), TRIGS(*), WORK(*)
SUBROUTINE CFFTS2_64(IOPT, N1, N2, SCALE, X, LDX, Y, LDY, TRIGS, IFAC, WORK, LWORK, IERR)
INTEGER*8 IOPT, N1, N2, LDX, LDY, IFAC(*), LWORK, IERR
COMPLEX X(LDX, *)
REAL SCALE, Y(LDY, *), TRIGS(*), WORK(*)
F95 INTERFACE
SUBROUTINE FFT2(IOPT, N1, [N2], [SCALE], X, [LDX], Y, [LDY], TRIGS,
& IFAC, WORK, [LWORK], IERR)
INTEGER*4, INTENT(IN) :: IOPT, N1
INTEGER*4, INTENT(IN), OPTIONAL :: N2, LDX, LDY, LWORK
REAL, INTENT(IN), OPTIONAL :: SCALE
COMPLEX, INTENT(IN), DIMENSION(:,:) :: X
REAL, INTENT(OUT), DIMENSION(:,:) :: Y
REAL, INTENT(INOUT), DIMENSION(:) :: TRIGS
INTEGER*4, INTENT(INOUT), DIMENSION(:) :: IFAC
REAL, INTENT(OUT), DIMENSION(:) :: WORK
INTEGER*4, INTENT(OUT) :: IERR
SUBROUTINE FFT2_64(IOPT, N1, [N2], [SCALE], X, [LDX], Y, [LDY], TRIGS, IFAC, WORK, [LWORK], IERR)
INTEGER(8), INTENT(IN) :: IOPT, N1
INTEGER(8), INTENT(IN), OPTIONAL :: N2, LDX, LDY, LWORK
REAL, INTENT(IN), OPTIONAL :: SCALE
COMPLEX, INTENT(IN), DIMENSION(:,:) :: X
REAL, INTENT(OUT), DIMENSION(:,:) :: Y
REAL, INTENT(INOUT), DIMENSION(:) :: TRIGS
INTEGER(8), INTENT(INOUT), DIMENSION(:) :: IFAC
REAL, INTENT(OUT), DIMENSION(:) :: WORK
INTEGER(8), INTENT(OUT) :: IERR
C INTERFACE
#include <sunperf.h>
void cffts2_ (int *iopt, int *n1, int *n2, float *scale, complex *x,
int *ldx, float *y, int *ldy, float *trigs, int *ifac, float
*work, int *lwork, int *ierr);
void cffts2_64_ (long *iopt, long *n1, long *n2, float *scale, complex
*x, long *ldx, float *y, long *ldy, float *trigs, long *ifac,
float *work, long *lwork, long *ierr);
PURPOSEcffts2 initializes the trigonometric weight and factor tables or com‐
putes the two-dimensional inverse Fast Fourier Transform of a two-
dimensional complex array. In computing the two-dimensional FFT, one-
dimensional FFTs are computed along the rows of the input array. One-
dimensional FFTs are then computed along the columns of the intermedi‐
ate results.
N1-1 N2-1
Y(k1,k2) = scale * SUM SUM W2*W1*X(j1,j2)
j1=0 j2=0
where
k1 ranges from 0 to N1-1 and k2 ranges from 0 to N2-1
i = sqrt(-1)
isign = 1 for inverse transform
W1 = exp(isign*i*j1*k1*2*pi/N1)
W2 = exp(isign*i*j2*k2*2*pi/N2)
In complex-to-real transform of length N1, the (N1/2+1) complex input
data points stored are the positive-frequency half of the spectrum of
the Discrete Fourier Transform. The other half can be obtained through
complex conjugation and therefore is not stored.
ARGUMENTS
IOPT (input)
Integer specifying the operation to be performed:
IOPT = 0 computes the trigonometric weight table and factor
table
IOPT = 1 computes inverse FFT
N1 (input)
Integer specifying length of the transform in the first
dimension. N1 is most efficient when it is a product of
small primes. N1 >= 0. Unchanged on exit.
N2 (input)
Integer specifying length of the transform in the second
dimension. N2 is most efficient when it is a product of
small primes. N2 >= 0. Unchanged on exit.
SCALE (input)
Real scalar by which transform results are scaled. Unchanged
on exit. SCALE is defaulted to 1.0 for F95 INTERFACE.
X (input) X is a complex array of dimensions (LDX, N2) that contains
input data to be transformed.
LDX (input)
Leading dimension of X. LDX >= (N1/2 + 1) Unchanged on exit.
Y (output)
Y is a real array of dimensions (LDY, N2) that contains the
transform results. X and Y can be the same array starting at
the same memory location, in which case the input data are
overwritten by their transform results. Otherwise, it is
assumed that there is no overlap between X and Y in memory.
LDY (input)
Leading dimension of Y. If X and Y are the same array, LDY =
2*LDX Else LDY >= 2*LDX and LDY must be even. Unchanged on
exit.
TRIGS (input/output)
Real array of length 2*(N1+N2) that contains the trigonomet‐
ric weights. The weights are computed when the routine is
called with IOPT = 0 and they are used in subsequent calls
when IOPT = 1. Unchanged on exit.
IFAC (input/output)
Integer array of dimension at least 2*128 that contains the
factors of N1 and N2. The factors are computed when the rou‐
tine is called with IOPT = 0 and they are used in subsequent
calls when IOPT = 1. Unchanged on exit.
WORK (workspace)
Real array of dimension at least MAX(N1,2*N2)*NCPUS, where
NCPUS is the number of threads used to execute the routine.
The user can also choose to have the routine allocate its own
workspace (see LWORK).
LWORK (input)
Integer specifying workspace size. If LWORK = 0, the routine
will allocate its own workspace.
IERR (output)
On exit, integer IERR has one of the following values:
0 = normal return
-1 = IOPT is not 0, 1
-2 = N1 < 0
-3 = N2 < 0
-4 = (LDX < N1/2+1)
-5 = LDY not equal 2*LDX when X and Y are same array
-6 = (LDY < 2*LDX or LDY odd) when X and Y are same array
-7 = (LWORK not equal 0) and (LWORK < MAX(N1,2*N2)*NCPUS)
-8 = memory allocation failed
SEE ALSO
fft
CAUTIONS
Y(N1+1:LDY,:) is used as scratch space. Upon returning, the original
contents of Y(N1+1:LDY,:) will be lost, whereas Y(1:N1,1:N2) contains
the transform results.
6 Mar 2009 cffts2(3P)