flatau@handel.CS.ColoState.Edu (flatau) (11/24/90)
I am looking for 3 dimensional Fourier transform
and/or 3 dimensional convolution program for complex
sequence of numbers.
I did my homework:
1. We are successfully using FOURN routine from "Numerical
recipes".
2. I've got Cliff Temperton's (of ECMWF) code, and since it has
jump and inc parameters was able to get 3 dimensional code
out of his 1D routines. Works nice and vectorizes well
on Cray (8-9 speedup over FOURN on Cray, but takes twice the memory).
3. I've noticed paper by Wells, N. H., Burrus, C. S. and Desobry/Boyer
" Three-dimensional Fourier conwolution with an array processor", Computers
in physics", Sep/Oct 1990 and requested their code.
4. Got NASA Ames (D. Bailey et consortes) stuff (didn't use it
much).
5. Checked Science Citation Index references. There is another
code published by A. Mobile, V. Roberto, F. Saitta,
Computer Physics Communications, 48, 1988, 313-318,
MFFT4: Four D FFT, but I don't have access to the source.
Still, there has to be some other software for multidimensional
convolutions and/or Fourier transforms with their own
virtues (e.g. vectorizing on Cray, in-place, etc.)
I would be willing to compare some of the all-FORTRAN
and {\it public domain}
routines and post the results on the net (unless you indicate otherwise).
The application
is for spatially invariant Green's function, and uses 64x64x64
matrices (max in this moment). I am using RISC6000/Cray-YMP/SUN Sparc
to run the code.
Cheers,
Peter