Newsgroups: comp.parallel.mpi
From: Christophe Koudella <ckoudell@ens-lyon.fr>
Subject: 3D_FFT_paralel
Organization: Ens Lyon
Date: Fri, 14 Mar 1997 22:39:48 +0100
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Message-ID: <3329C5A4.41C6@ens-lyon.fr>

Hello everybody, 

I am writing a Navier-Stokes code for Cray T3E, a pseudospectral 
code. 

Currently I am using the PSCFFT3D 3dimensional fast fourier 
transform routine implemented by Cray themselves. 

The performance of this routine is however very bad on T3E and I am 
thinking of writing my own 3 DIMENSIONAL FFT routine. 

I am using MPI or even SHMEM. Can you give me a clue as to how to do it, 
so that things get faster. If you got a routine or a skeleton, I would
be glad to have it. I am just begining with all this and am very 
scared of communications and the like. 

This could imply, as suggested to me, to cut the computational 
domain into slices , such that the data is distributed only 
in one direction. So a two dimensional fft may be done on a single 
processor straight away. Problem comes from the third direction. 

Hope you can help me , 

Thanks

Christophe KOUDELLA
ckoudell@ens-lyon.fr