Path: utzoo!utgpu!news-server.csri.toronto.edu!helios.physics.utoronto.ca!ists!yunexus!oz
From: oz@yunexus.yorku.ca (Ozan Yigit)
Newsgroups: comp.lang.c
Subject: Re: Some optimization results
Keywords: trivial pursuit
Message-ID: <22723@yunexus.YorkU.CA>
Date: 14 May 91 18:57:33 GMT
References: <22683@yunexus.YorkU.CA> <5850:May901:25:0491@kramden.acf.nyu.edu> <3501@charon.cwi.nl> <14077:May1321:14:2691@kramden.acf.nyu.edu>
Sender: news@yunexus.YorkU.CA
Organization: York U. Communications Research & Development
Lines: 25

Dan Bernstein writes:

>On any given bottleneck, someone could either do the obvious big
>optimizations and stop, or do every optimization that he can see that
>will gain a few percent.

An unreliable, non-universal few percent is almost never worth the
unreadable, unmaintainable mess that is produced as an end result of
"every optimization", given a sufficiently well-written program to
start with.

>I can keep pointing out how hand optimization gives huge speedups.

 wollman@emily.uvm.edu (Garrett Wollman) points out elsewhere:

 | [For example, yabba will not work for (some subset of the set of)
 | large binary files on our SGI 4D/340S.  Dan also spent so much effort
 | unrolling loops that the source is nearly incomprehensible, and breaks
 | a parallelizing preprocessor which can be used by the MIPS/SGI
 | compiler to great effect.]

... oz
---
In seeking the unattainable, simplicity | Internet: oz@nexus.yorku.ca
only gets in the way. -- Alan J. Perlis | Uucp: utai/utzoo!yunexus!oz