Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!swrinde!cs.utexas.edu!sun-barr!rutgers!cmcl2!kramden.acf.nyu.edu!brnstnd
From: brnstnd@kramden.acf.nyu.edu (Dan Bernstein)
Newsgroups: comp.lang.c
Subject: Re: Some optimization results
Message-ID: <3072:May700:46:3691@kramden.acf.nyu.edu>
Date: 7 May 91 00:46:36 GMT
References: <444@smds.UUCP> <186@shasta.Stanford.EDU>
Organization: IR
Lines: 25

In article <186@shasta.Stanford.EDU> shap@shasta.Stanford.EDU (shap) writes:
> >(a)	Make arrays static or malloced; don't make them automatic.
> Don't bother. Get a serious compiler.

``> (a) Write fast programs.'' ``Don't bother. Get a Cray.''

Unfortunately, buying a Cray will only help code that is never used on
another machine. Similarly, getting a better compiler will only help
code that is distributed in binary form for that machine, or is not
distributed at all.

I find both pieces of advice remarkably counterproductive, as they take
time away from serious thought about optimization.

> More to the point, use the memcmp() or bcmp() routines.  They are probably
> better suited to the problem because they are hand-tuned to the system
> at hand.

Piercarlo's study some time back proved otherwise.

> In general, the compiler is probably good at unrolling.

Most compilers can't unroll anything.

---Dan