Yes, what Ross says is essentially true. I support FFTW and old and new sgi fft's (at least through 9, have not played with an sgi in a very long time now...). The FFTW fft's are maybe 10% faster on a uniprocessor pentium 4, but also prone to producing slightly variable results due to variable algorithm selection (fftw has this feature whereby it can select the best algorithm for a given machine; the answer is sometimes slightly random, so you get the fft's done 2 slightly different ways; a bit of a nuisance). The fftw stuff also supports a prime factor of 7, but I am not sure that really helps much in most scenarios. So I do sort of favor the pubfft stuff we have, which started with netlib stuff that Tom Darden picked up. I sat down and did a bit of a rewrite of all the code and amazingly managed to make it even faster. So our own fft's are actually really pretty darn good, compile fast, and it is really is easy to find and link to the library (there is none ;-)).
I COULD imagine that IBM may have something special for some screwball architecture or other that might be faster...
Regards - Bob
----- Original Message -----
From: Ross Walker
To: amber-developers.scripps.edu
Sent: Friday, March 07, 2008 4:45 PM
Subject: RE: amber-developers: ffts
Hi Carlos,
We use our own FFT routines (from pub fft) and so do not need to link to any FFT libraries. Bob supports linking to things like FFTW with PMEMD but his own custom FFT routines are normally just as good if not better.
All the best
Ross
/\
\/
|\oss Walker
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
|
http://www.rosswalker.co.uk | PGP Key available on request |
Note: Electronic Mail is not secure, has no guarantee of delivery, may not be read every day, and should not be used for urgent or sensitive issues.
----------------------------------------------------------------------------
From: owner-amber-developers.scripps.edu [mailto:owner-amber-developers.scripps.edu] On Behalf Of Carlos Simmerling
Sent: Friday, March 07, 2008 13:35
To: amber-developers.scripps.edu
Subject: amber-developers: ffts
I've been asked what kind of fft libraries we want to have access to,
and I'm not sure what we're using. can someone tell me what we do now?
sorry to be naive about this
carlos
--
===================================================================
Carlos L. Simmerling, Ph.D.
Associate Professor Phone: (631) 632-1336
Center for Structural Biology Fax: (631) 632-1555
CMM Bldg, Room G80
Stony Brook University E-mail: carlos.simmerling.gmail.com
Stony Brook, NY 11794-5115 Web: http://comp.chem.sunysb.edu
===================================================================
Received on Sun Mar 09 2008 - 06:07:38 PDT
