From: "John S. Fine" <johnfine AT erols DOT com>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Optimizations
Date: Tue, 29 Sep 1998 10:13:13 -0400
Lines: 27
Message-ID: <3610EAF9.A58@erols.com>
References: <Pine DOT SUN DOT 3 DOT 91 DOT 980923191310 DOT 2968I-100000 AT is> <360996D2 DOT 382F AT erols DOT com> <3610272A DOT E54C23F7 AT mailexcite DOT com>
Reply-To: johnfine AT erols DOT com
NNTP-Posting-Host: 207-172-240-144.s17.as3.bsd.erols.com
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Precedence: bulk

Doug Gale wrote:

> What are you making that requires such extreme performance that you are willing
> to sift through the assembler and make sure it sqeezes every drop of speed from
> the machine? If your application is SO performance sensitive, why target it for
> a junky old 486 with no L2 cache? :)

  The "junky old 486 with no L2 cache" is the hand held computer that
my employer manufatures.  It is (very well) designed for ruggedness
and long battery life, not for CPU power.  By writing the frequently
used routines in assembler, I achieve very good performance.  In this
project I needed to have a set routines in portable C for running the
same routines on other platforms.  For maintenance reasons I wanted
to have just one version of the source code without a lot of conditional
compilation.  After trying it that way, I gave up on the idea of single
source code and switched back to writing the 486 version in assembler.

> Besides, algorithmic optimizations will always make a bigger improvement than
> low-level optimizations anyway. As the old saying goes: the fastest instruction
> is the one you never execute!

  I think I got the algorithm optimized before I started worrying about
the low-level optimizations.  

-- 
http://www.erols.com/johnfine/
http://www.geocities.com/SiliconValley/Peaks/8600/