Path: utzoo!attcan!uunet!lll-winken!lll-tis!ames!mailrus!tut.cis.ohio-state.edu!rutgers!cmcl2!polyof!kepler1!rjfrey
From: rjfrey@kepler1.UUCP (Robert J Frey)
Newsgroups: comp.lang.apl
Subject: Re: APL for Parallel Algorithms
Keywords: APL, parallel processing
Message-ID: <45@kepler1.UUCP>
Date: 9 Dec 88 18:25:36 GMT
References: <3779@hubcap.UUCP> <1522@plx.UUCP>
Reply-To: rjfrey@kepler1.UUCP (Robert J Frey)
Organization: Kepler Financial Mgmt, Setauket, NY
Lines: 32

In article <1522@plx.UUCP> evan@plx.UUCP (Evan Bigall) writes:
>
>I was inolved in the implementation of a parallel/vector APL compiler at IBM's 
>Watson Labs.  One of the largest problems with parallel APL is that to really
>get a speed up, you need a compiler and we all know how much fun compiling 
>APL is...
>

I certainly would be interested in any references you could make available.
Perhaps, in terms of APL, a more modest goal would be appropriate:  Instead
of thinking of APL as a means for specifying  or implementing parallel
algorithms, just try to implement APL in a way in which allows it to exploit
the vector/parallel facilities of the target environment.  For example, if
I want to compute the inner product A+.xB given the simple numeric matrices
A and B, this seems easy to implement in a way which exploits vector/
parallel facilities.  Another example would be many uses of the "for each"
operator; doesn't this essentially specify independent execution chains?

I can recall one instance at a former employer where we took a "linear algebra
intensive" application in APL2 and quadNA'd some ESSL functions to handle
such things as inner products, pivots, etc. and realized a thirty-fold increase
in speed (from ~5 min to ~9 sec.).  If more of this sort of thing can be done
in ways more transparent to the user, then APL would be much improved even
though it may not represent a general solution to specifying parallel 
algorithms.

==============================================================================
|Dr. Robert J. Frey               | {acsm, sbcs, polyof}!kepler1!rjfrey      |
|Kepler Financial Management, Ltd.|------------------------------------------|
|100 North Country Rd., Bldg. B   | The views expressed are wholly my own and|
|Setauket, NY  11766              | and do not reflect those of the Indepen- |
|(516) 689-6300 x.16              | dent Republic of Latvia.                 |
==============================================================================