Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!sdd.hp.com!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!emory!hubcap!pjn
From: pjn@cs.UMD.EDU (P. J. Narayanan)
Newsgroups: comp.parallel
Subject: Re: CM-2 SIMD question
Message-ID: <12519@hubcap.clemson.edu>
Date: 8 Jan 91 16:31:13 GMT
Sender: fpst@hubcap.clemson.edu
Reply-To: pjn@cs.umd.edu
Lines: 33
Approved: parallel@hubcap.clemson.edu

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

	I am afraid there isn't really a better way to do it, if the
individual methods to compute equations of state are quite different,
as you seem to imply. Each PE of the Connection Machine can
communicate independently with any other PE. There is a trick by which
each can address its own memory (an array in it) independently of one
another also. But all PEs are constrained to do identical operations
(add/subtract/memory read etc), or none at all, at all times.

	If the computations are not *very* different from each other,
you may be able to "parametrize" them into a table of values and bring
the computation into a pure SIMD framework. For instance, compute all
logical functions using AND and NOT alone, with the right values taken
from a table and in the right order. Similarly, subtraction can be
avoided by using a multiplication by a table value and addition -- use
1 as the table value to get addition, and -1 to get subtraction etc. I
don't think anything more complicated than toy-programs can be
transformed efficiently this way.

<John K. Prentice

P J Narayanan