From:	leathm AT solwarra DOT gbrmpa DOT gov DOT au (Leath Muller)
Message-Id:	<199702060351.NAA13273@solwarra.gbrmpa.gov.au>
Subject:	Re: doubles vs. floats
To:	junk AT defeating DOT email DOT address
Date:	Thu, 6 Feb 1997 13:51:45 +1000 (EST)
Cc:	djgpp AT delorie DOT com
In-Reply-To:	<KyW5HCAANE+yEwij@foobar.co.uk> from "Paul Shirley" at Feb 5, 97 08:22:24 am

> > On pre-pentium, the conversion takes time. :) 
 
> NOT AS FAR AS THE USER KNOWS! (Can you tell I'm getting annoyed?)

Geez, chill out, why get annoyed over a thread like this one? ;)
Anway, who care's if the user knows, I was talking from a _cycle_
level point of view.
 
> On 8087,80287 and 80387 using a 32 bit float is *ALWAYS* faster than a
> 64 bit double. Check the bloody manual.

What manual? Why? I have a pentium, write for a pentium, optimise for a
pentium - why would I want to get a manual for old technology? The pentium
is a huge leap over all previous x86 chips for performance...
 
> On a 486/7 a float is the same speed in theory, in practice it will be
> *faster* than a double because of caching/bus width effects.

Yeah, but I'm not talking 486 - cause I don't care about 486...its not
my fault its not worth supporting anymore... :) On a pentium, as long
as everything is aligned properly, which DJGPP will do for you, caching
and bus width effects can be made to be negligble. And IF your optimising
properly and thoroughly, you will optimise to avoid cache misses, and
thus it shouldn't be a problem!!  SOOOO, stop telling about 486's and
487's, I don't care! I don't mean to insult anyone with that statement,
its just for high speed 3D graphics applications, anything pre-pentium just
can't cut it... NOW can we kill this thread?  Just one thing: What CPU
do you have?  :)

Leathal.

- Raw text -