Path: utzoo!mnetor!uunet!husc6!bloom-beacon!gatech!mcnc!ecsvax!urjlew
From: urjlew@ecsvax.UUCP (Rostyk Lewyckyj)
Newsgroups: comp.arch
Subject: Re: Powers of 2, was Re: RISC is a nasty no-no!
Message-ID: <4734@ecsvax.UUCP>
Date: 5 Mar 88 06:00:42 GMT
References: <17415@think.UUCP> <1339@alliant.Alliant.COM> <24529@yale-celray.yale.UUCP>
Organization: UNC Educational Computing Service
Lines: 20
Summary: Just some details


There have been several comments about how problems dealing with
power of two points are a no no on many computers because they
lead to memory bank access conflicts. There has also already been
published a fairly simple work around. 
Don't declare your arrays with power of two shape factors.
i.e don't DIMENSION any component to be a power of two.
There is usually nothing which forces your dimensions to be 
exactly the size of your problem. If you are doing an FFT of
256 or 1024 points, just dimension the array to be 257 or 1025
elements long.
I believe that on the CRAY the magic number for bank conflicts
is something like 48 which is not a power of two.
On the IBM 3090 16 is a magic number because of the sructure
of the cache rather than memory bank interleave.
-----------------------------------------------
  Reply-To:  Rostyslaw Jarema Lewyckyj
             urjlew@ecsvax.UUCP ,  urjlew@tucc.bitnet
       or    urjlew@tucc.tucc.edu    (ARPA,SURA,NSF etc. internet)
       tel.  (919)-962-9107