Date: Wed, 5 Jun 1996 17:27:42 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: mlarch AT bigdog DOT fred DOT net
Cc: djgpp AT delorie DOT com
Subject: Re: [?] aligning value returned by calloc/malloc
In-Reply-To: <199606051209.IAA22861@bigdog.fred.net>
Message-Id: <Pine.SUN.3.91.960605172417.26657E-100000@is>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII


On Wed, 5 Jun 1996 mlarch AT bigdog DOT fred DOT net wrote:

> I was working on speeding up a spaning tree algorithm, large chunks 
> of binary sceen data, using djgpp's gcc and assembler. Trying to stream
> line the process for max-speed, it would make some of the loop code
> much less thick and much more easy to keep in the cache if I could
> count on a certain data being aligned in certian ways.

OK.  Just be sure that the part of the program that you are trying to 
speed up indeed takes a significant percent of the CPU time (use the 
profiler).  And beware: the Intel on-chip cache can surprise you wrt the 
effects of such optimizations: I have found that many optimizations that 
should have improved the performance indeed do that when I disable the 
cache, but not when the cache is on.