Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!zaphod.mps.ohio-state.edu!casbah.acns.nwu.edu!ucsd!network.ucsd.edu!jacobi.ucsd.edu!mbk
From: mbk@jacobi.ucsd.edu (Matt Kennel)
Newsgroups: comp.arch
Subject: Re: Computers for users not programmers
Message-ID: <4612@network.ucsd.edu>
Date: 10 Feb 91 02:21:06 GMT
References: <27AF17B9.72E2@tct.uucp> <5275@mentor.cc.purdue.edu> <27B19A39.321E@tct.uucp>
Sender: news@network.ucsd.edu
Organization: Intstitute for Nonlinear Science, UCSD
Lines: 46
Nntp-Posting-Host: jacobi.ucsd.edu

In article <27B19A39.321E@tct.uucp> chip@tct.uucp (Chip Salzenberg) writes:
>According to hrubin@pop.stat.purdue.edu (Herman Rubin):
>>There is the mistaken view that hardware should be designed to particular
>>languages, and never mind that some programs may be many times slower
>>because of the lack of particular instructions.

>Machines have been designed for efficient execution in the most common
>cases.  If the most common cases are compiled C and Fortran programs,
>optimizing the hardware for those cases is only natural.
>
>Remember, Herman, your instruction mix is radically atypical.

I think Mr. Rubin's "beef" arises from a difference in computing cultures.
As far as I can discern from his previous postings, he regrets the lack of
both hardware and software provisions for mixed-mode and multi-precision
integer arithmetic, which he claims to be "trivial" to implement.  I have no
idea if this is true, but it certainly seems likely compared to the massive
effort that designers put into elaborate cache architectures and
multi-processor provisions.  This is the result of project managers at
successful CPU design firms optimizing their resources (i.e. employees' labor)
for the common case.

He writes computer programs to do _mathematics_, whereas the remaining
99% of the numerical programming world writes programs to do _science
and engineering_, where standard floating-point operations are the norm
and hence the object of substantial effort on the part of computer designers.

The only other group I can think of that does large-scale
computing similar to Mr Rubin's is the NSA, in the cryptanlysis field.
Unfortunately for the general public, they have the resources to acquire
their own customized computers, and the power to keep them an exclusive 
secret.  Other than them, number theorists and other mathemeticians
are a 'trivial' market, which is quite unfortunate if you happen to be one.

Mr Rubin's other complaint, that computer languages do not take advantage of
available hardware is also true.  Before you write a rebuttal that says "but
hardly anybody wants to do that stuff anyway", consider vector and matrix
operations.  They're not an intrinsic part of any commonly used language (if
you discount the vapor-Fortran-9X-X-X) but nobody would dare say that
they're not very useful, considering there is significant hardware
provisions in many large computers for accelerated vector operations.  (And
I still need to be convinced that C++ can do them efficiently for both
scalar and vector machines).

Matt K
mbk@inls1.ucsd.edu