Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!gem.mps.ohio-state.edu!uakari.primate.wisc.edu!dogie.macc.wisc.edu!schaefer!wayne
From: wayne@schaefer.MATH.WISC.EDU (Rick Wayne)
Newsgroups: comp.object
Subject: Re: Parallelism and OOPSs/Number crunching
Message-ID: <480@schaefer.MATH.WISC.EDU>
Date: 9 Oct 89 20:46:04 GMT
References: <477@schaefer.MATH.WISC.EDU> <4110@hplabsz.HPL.HP.COM> <GDA.89Oct9114712@creare.creare.UUCP>
Reply-To: wayne@schaefer (Rick Wayne)
Organization: Mathematics Department, University of Wisconsin - Madison
Lines: 20

In article <GDA.89Oct9114712@creare.creare.UUCP> gda@creare.creare.UUCP (Gray Abbott) writes:
>
>I'm especially interested
>in the possibility of using C++ to "hide" the vectorization of
>computations, so that the programs are more transparent, without
>sacrificing too much in performance (or even improving performance
>by increasing the amount of code that is effectively vectorized).

this is what i was trying to get at in the original posting.  (thanks for
the education on the way, folks!)

i was confusing distribution with parallelism; sloppy thinking on my part.
however, i keep wondering that if you solve the synchronization problem 
inherent in distributed code, might you not get parallelism for free?
actually, i guess it would probably be the other way around -- make your
code parallel, and the rendezvous-ing involved would obviate explicit
synchronization.  so the programmer still has to sweat decomposition
and vectorizing if the compiler doesn't.

rw