Message-ID: <19990815025507.23694@atrey.karlin.mff.cuni.cz>
Date: Sun, 15 Aug 1999 02:55:07 +0200
From: Jan Hubicka <hubicka AT atrey DOT karlin DOT mff DOT cuni DOT cz>
To: pgcc AT delorie DOT com
Subject: Re: optimizing for k6
References: <3 DOT 0 DOT 32 DOT 19990814040832 DOT 01181ec0 AT pop DOT xs4all DOT nl> <19990814183125 DOT 24893 AT atrey DOT karlin DOT mff DOT cuni DOT cz> <19990814214330 DOT A3812 AT win DOT tue DOT nl>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
X-Mailer: Mutt 0.84
In-Reply-To: <19990814214330.A3812@win.tue.nl>; from Ronald de Man on Sat, Aug 14, 1999 at 09:43:30PM +0200
Reply-To: pgcc AT delorie DOT com
X-Mailing-List: pgcc AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

> On Sat, Aug 14, 1999 at 06:31:25PM +0200, Jan Hubicka wrote:
> > Isn't that mainly because of memory consumed by your program has decreased
> > when you changed your datastructure? K6 is very sensitive about memory,
> > because it have quite small caches and refills are more costy than on the
> > Intel CPU familly.
> 
> If I'm not mistaken, the K6 is advertised to have 32k+32k L1 cache,
> while CPU's by Intel mostly have 16k+16k. So your remark seems to
> imply that these numbers are not comparable?
Don't know. I am getting more problems with cache on AMD.
The instruction cache have lots of predecode information (5 bits as I remember)
and loading of caches seems to cause more stalls. I didn't looked exactly
for the reason, just I've noticed that shortening code/reordering loops
to be cache friendly (reading videoram line by line) makes huge speedups
on K6 while smallers on my Pentium.
I was trying to make extreme cache alignment (64kb) the loss in performance
on AMD was roughly 60% while on pentium 15%.
Maybe something is just worng with my setup, but loading cache line seems
to cause many cycles stall.
I am using 100Mhz memories here while slow ones on Pentium...

Honza
> 
> Ronald

-- 
                       OK. Lets make a signature file.
+-------------------------------------------------------------------------+
|        Jan Hubicka (Jan Hubi\v{c}ka in TeX) hubicka AT freesoft DOT cz         |
|         Czech free software foundation: http://www.freesoft.cz          |
|AA project - the new way for computer graphics - http://www.ta.jcu.cz/aa |
|  homepage: http://www.paru.cas.cz/~hubicka/, games koules, Xonix, fast  |
|  fractal zoomer XaoS, index of Czech GNU/Linux/UN*X documentation etc.  | 
+-------------------------------------------------------------------------+