Date:	Thu, 20 May 1999 20:38:52 +0200 (MET DST)
From:	Henrik Berglund SdU <adb94hbd AT mds DOT mdh DOT se>
X-Sender:	adb94hbd AT legolas DOT mdh DOT se
To:	pgcc AT delorie DOT com
Subject:	Re: [Re: K6 & i686 ASM code]
In-Reply-To:	<19990520152514.13578.qmail@www0i.netaddress.usa.net>
Message-ID:	<Pine.GSO.4.05.9905202023120.5930-100000@legolas.mdh.se>
MIME-Version:	1.0
Reply-To:	pgcc AT delorie DOT com
X-Mailing-List:	pgcc AT delorie DOT com
X-Unsubscribes-To:	listserv AT delorie DOT com

On 20 May 1999, ADAM SCHROTENBOER wrote:

> True, but you still haven't really answered my q. I guess that I can't use
> i686 ASM, but maybe i586 ASM for zlib. Or maybe just use -march=k6. But what's
> better, -march=k6, or the i586 ASM. 
If you want to use ASM i think you should use i585 since i686 might use
unsupported instructions. I have benchmarked with bytemark and -march=k6
is faster than -march=i586 so compiler optimisation might be faster than
hand made i586 asm for k6 but dont think so. the best thing is to try them
all and compare the speed.

> For the kernel, it's even more of a q. I would once again assume I shouldn't
> use i686 ASM, 
it might not be stabel. in the kernel config it says one should use i586
for k6/k6-2.

>but then is there clean C code for the code in the ASM portions?
> If so, how do I use that instead? Is it advantageous to use the C code,
> compiled for K6, or to just use the 586 ASM code?
You dont gain so much by optimising the kernel since most heavy
calculations are in the programs. You can use i586 and increase the
compiler optimisation to -O6, not that i think you will be able to se a
difference.




-----------------------------------------------------------------------------
Henrik DOT Berglund AT mds DOT mdh DOT se 
http://www.mds.mdh.se/~adb94hbd/

- Raw text -