Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!elroy.jpl.nasa.gov!ncar!gatech!hubcap!fpst
From: kelly@think.com (Kelly Pickard)
Newsgroups: comp.parallel
Subject: Spectral Code ideas
Message-ID: <1991May22.200828.13278@hubcap.clemson.edu>
Date: 22 May 91 15:19:40 GMT
Sender: kelly@quake.think.com
Reply-To: Kelly Pickard <kelly@think.com>
Organization: Clemson University
Lines: 38
Approved: parallel@hubcap.clemson.edu



  I forwarded the article to one of the people I
  work with for his comments, listed below.  Additionally,
  TMC is working on a pentadiagonal solver that will be
  included in a future release of the CMSSL (CM Scientific
  Software Library).

  Regards,
    Kelly Pickard
    Thinking Machines Customer Support

------

   Regarding this spectral problem, I have no experience with 
   penta-diagonal solvers, but it seems to me that the calculation 
   would be order nx*ny, whereas the FFT in nx*ny*ln(ny). This
   seems to indicate that the FFT will dominate his calculations.
   He should try weighting the y axis so his fft is mostly on 
   processor, and compare that to the generic layout performance.

							Rich

   ------- Forwarded Message

   Date:    Fri, 17 May 91 15:02:36 -0400 
   From:    kurt@Think.COM
   To:      science@Think.COM
   Subject: spectral code request


   from comp.parallel:
   Article 2604 (1 more) in comp.parallel (moderated):
   From: jek@cs.duke.edu (James E. Kittock)
   Subject: Pseudospectral codes on the CM-II...
   Date: 16 May 91 18:50:50 GMT

....