Xref: utzoo comp.lang.c:8610 comp.lang.misc:1345
Path: utzoo!mnetor!uunet!husc6!tut.cis.ohio-state.edu!bloom-beacon!mit-eddie!bbn!rochester!PT.CS.CMU.EDU!IUS1.CS.CMU.EDU!edw
From: edw@IUS1.CS.CMU.EDU (Eddie Wyatt)
Newsgroups: comp.lang.c,comp.lang.misc
Subject: Re: Languages vs. machines (was Re: The need for D-scussion)
Message-ID: <1226@PT.CS.CMU.EDU>
Date: 25 Mar 88 18:52:21 GMT
References: <12176@brl-adm.ARPA> <1988Mar11.215238.976@utzoo.uucp> <5308@hall.cray.com>
Sender: netnews@PT.CS.CMU.EDU
Organization: Carnegie-Mellon University, CS/RI
Lines: 40

This really doesn't belong in comp.lang.c but.....

> Compilers can
> hide lots of annoying differences, but they can't change a nifty linked
> VAX-optimal data structure into a simple vectorizable Cray-optimal array.

   I ask you to prove the above statement.  

   I think this debate is really  over who is responsible for optimizing
code.  The compiler or the programmer.  It's the comilers job in my
opinion.   I'm not going to list the many reasons why.  However,
I have to admit I do hand tune my code sometimes to the paricular machine
that I'm working on, but only because the compiler I use "ccp" is
not a very good optimizing compiler.

> This is not to say that performance goals necessarily rule out portability,
> just that portability may be restricted to a smaller range of systems.
> Chris' hypothetical 1TFLOP, 10TFLOP, and 100TFLOP machines (see your
> local Mimsy Data sales office :-)) are not likely to be much like
> an 8088 or 6502.

   You laugh, a representative of IMB Yorktown came to CMU and gave
a talk on the TF3, a 3 teraflop machine that they plan to deliver in
two years.  It consists of 4096 processing elements interconnected
by an omega ring.  BTW, if I remeber correctly, the machine will
take on the order of 30 Megawatts of power to run.  Does anyone know
how much money that is per hour?

> (Could a language be designed that would preserve some measure of portability
>  across some set of supercomputer systems while allowing maximal control and
>  performance? C and C++ are great at this amongst scalar byte-addressable
>  architectures but seem inadequate (to me!) to the task of extracting the most
>  speed from a multiprocessor vector box with only large-word addressing.
>  I've been trying (on and off) to devise such a language.  Not easy, and probably
>  not desirable, but an interesting experiment nontheless.  Mail ideas, please.)

  How about data-flow languages.
-- 

Eddie Wyatt 				e-mail: edw@ius1.cs.cmu.edu