Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!emory!hubcap!"John
From: mccalpin@perelandra.cms.udel.edu (John D. McCalpin)
Newsgroups: comp.parallel
Subject: Re: CM-2 SIMD question
Message-ID: <12532@hubcap.clemson.edu>
Date: 9 Jan 91 12:44:08 GMT
References: <12515@hubcap.clemson.edu>
Sender: fpst@hubcap.clemson.edu
Organization: College of Marine Studies, U. Del.
Lines: 38
Approved: parallel@hubcap.clemson.edu

> On 8 Jan 91 13:49:05 GMT, john@ghostwheel.unm.edu (John Prentice) said:

John> I have a SIMD question motivated by the CM-2.  We are doing
John> finite difference calculations on a simple structured grid.
John> [....]  However, different nodes in our mesh use different
John> equations of state.  Some are tabular, some are analytic, some
John> are iterative.  Clearly you can't perform all these calculations
John> in lock step.  So, I assume we have to turn off all processors
John> except those using a specific equation of state, do the EOS
John> calculations for those nodes, and then turn those processors off
John> and turn on those for the next EOS, etc... until we have
John> exhausted all the types of EOS we need.  Is there a better way
John> to do this?

Not easily!

I talked with Danny Hillis about this last year.  In his original
plan, there were to be multiple instruction streams operating
simultaneously.  Each processor would have a few bits telling it which
instruction stream to process.  In the end they decided that most
cases could be handled by a simple masking bit, so that is what got
put it ---- I guess it was a lot cheaper and it had the advantage of
not requiring any more bandwidth from the front end. 

I was interested at the time because of the possibility of
implementing boundary condition code using alternate instruction
streams, but we also had users who had locally changing physics (cloud
modellers) who needed essentially the same thing that you are
discussing....

I still think that having several instruction streams (4 or 8 ?) would
greatly increase the flexibility of the machine.  It would allow a lot
of the benefits of MIMD processing with all the advantages of the SIMD
architecture....
--
John D. McCalpin			mccalpin@perelandra.cms.udel.edu
Assistant Professor			mccalpin@brahms.udel.edu
College of Marine Studies, U. Del.	J.MCCALPIN/OMNET