Comer Duncan wrote:
>
>I am working on a problem for which I need a 2d fft of a real
>array, do some work on the tranformed array, solve some algebraic
>equations for the key variables, and then perform the inverse
>fft to form the updated variable [this is a incompressible flow
>solver]. I am about to write the thing myself and thought I would
>ask the community about fft routines before I make the final choices
>on storage, real vs complex arrays, etc. The machine on which I will
>be computing is an Origin2K running IRIX 6.5.xxx and using the
>SGI MipsPro 7.2.x compilers.
On earlier versions of IRIX and Compilers I once found that the routines
of the public domain package FFTW were more efficient for my application than
the once in complib.sgimath.a.
FFTW can be found at:
http://theory.lcs.mit.edu/~fftw/
>My question: Does anyone know whether the fft routines that SGI
>provides are in fact highly optimized? If not, what _is_ the best
>fft routine you know about as far as speed is concerned? The vendor?
>I also have access to a Cray T94 and thus have the same question
>for that platform.
I do not know because my experience is the performance of FFT routines
can be very platform dependent.
You will hace to make a principle decision:
1) make use of the best routine for the platform and maintain different
code for different machines
2) use the overall best FFT and just maintain one code set for all
platforms
>Of course, I am assuming that the SGI supplied fft routines are f90
>routines, but quite frankly it looks as though this may not be the
>case....I note that the man page for fft mentions fortran but not
>fortran90 specifically. If these are fortran 77 but not fortran 90
>routines, is there a problem?
What I remember (but may be I'm wrong) the routines in complib.sgimath are
basically the routines from netlib's fftpack.
>Thanks for replies from those fft'ers who have some knowledge about
>efficient f90 bindings, if these routines exist...
I created a module which can be used (with conditional compilation
(C-preprocessor) ) to calculate FFT's with either:
FFTPACK
ESSL (IBM's AIX)
FFTW
DXML (FOR DEC's VMS ) (this will work with a small change also for Digital
Unix)
SGIMATH (SGI's IRIX)
At the moment only 2D DoubleComplex - DoubleComplex FFT and inverse FFT's are
implemented (I hardly use others) but it is obvious how to do the other
transforms.
If you are interested I can send you the module.
Jouk
Comer Duncan
------------
Professor of Physics
Department of Physics and Astronomy
Bowling Green State University
Bowling Green, OH 43403
email: [log in to unmask]
phone: (419) 372 8108
fax:(419) 372 9938
------------
================== RFC 822 Headers ==================
Return-Path: [log in to unmask]
Received: by dutsm1120.stm.tudelft.nl (UCX V4.2-21A, OpenVMS V7.1-1H2 Alpha);
Wed, 31 Mar 1999 05:45:36 +0200
Received: from ns2.tudelft.nl by dutsm43.stm.tudelft.nl (AIX 3.2/UCB 5.64/4.03)
id AA25057; Wed, 31 Mar 1999 05:45:40 +0200
Received: from mailout1.mailbase.ac.uk by mailhost1.tudelft.nl
(PMDF V5.1-12 #D3520) with ESMTP id <[log in to unmask]> for
[log in to unmask]; Wed, 31 Mar 1999 05:45:30 +0200 (MET DST)
Received: from naga.mailbase.ac.uk (naga.mailbase.ac.uk [128.240.226.3])
by mailout1.mailbase.ac.uk (8.9.1a/8.9.1) with ESMTP id EAA15431; Wed,
31 Mar 1999 04:45:15 +0100 (BST)
Received: (from daemon@localhost) by naga.mailbase.ac.uk (8.8.x/Mailbase)
id EAA14581; Wed, 31 Mar 1999 04:41:53 +0100 (BST)
Received: from chandra.bgsu.edu (chandra.bgsu.edu [129.1.9.146])
by naga.mailbase.ac.uk (8.8.x/Mailbase) with ESMTP id EAA14569; Wed,
31 Mar 1999 04:41:49 +0100 (BST)
Received: (from gcd@localhost)
by chandra.bgsu.edu (980427.SGI.8.8.8/970903.SGI.AUTOCF) id WAA13803; Tue,
30 Mar 1999 22:10:56 -0500 (EST)
Date: Tue, 30 Mar 1999 22:10:56 -0500 (EST)
From: [log in to unmask] (Comer Duncan)
Subject: efficient 2d fft routines?
Sender: [log in to unmask]
To: [log in to unmask]
Cc: [log in to unmask]
Errors-To: [log in to unmask]
Reply-To: [log in to unmask] (Comer Duncan)
Message-Id: <[log in to unmask]>
Precedence: list
X-Unsub: To leave, send text 'leave comp-fortran-90' to [log in to unmask]
X-List-Unsubscribe:
<mailto:[log in to unmask]>
X-List: [log in to unmask]
Ceterum censeo tertium millennium post Christum natum anno MMI incepturum esse
>------------------------------------------------------------------------------<
Jouk Jansen
[log in to unmask]
Technische Universiteit Delft tttttttttt uu uu ddddddd
Nationaal centrum voor HREM tttttttttt uu uu dd dd
Rotterdamseweg 137 tt uu uu dd dd
2628 AL Delft tt uu uu dd dd
Nederland tt uu uu dd dd
tel. 31-15-2781536 tt uuuuuuu ddddddd
>------------------------------------------------------------------------------<
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|