Subject: Re: gcc on x86 - any cool gimmicks?
To: David Brownlee <abs@netbsd.org>
From: Rasputin <rasputin@idoru.mine.nu>
List: port-i386
Date: 08/08/2002 11:40:26
* David Brownlee <abs@netbsd.org> [020808 10:51]:
> On Wed, 7 Aug 2002, Aaron J. Grier wrote:
> 
> > On Wed, Aug 07, 2002 at 02:58:11PM -0500, Peter Seebach wrote:
> > > It just occurred to me:  By default, gcc is building 386-compatible
> > > binaries with reasonably general optimization.  Does gcc know enough
> > > to optimize well for PII or PIII processors?  Is doing this remotely
> > > safe or supported?
> >
> > pkgsrc/devel/cpuflags.  :)
> 
> 	Just to chime in on that :) - I've been using cpuflags to
> 	optimise kernel, userland, and all packages (up to kde3
> 	and openoffice) on an assortment of boxes including:
> 
> 	K6-2/PII/PIII/sparc10/sparc5/sparc20/ultra1/shark/DEC 433au.
> 
> 	The only problem I've seen is on NetBSD/sparc older compilers
> 	(NetBSD 1.4 and maybe 1.5) on compiling the kernel
> 	-march=supersparc on _some_ boxes. Of course the boxes are
> 	running perfectly with cpuflags under 1.6_BETA5 now, which
> 	makes tracking it down a little harder :)

'Me too' - my SS20 refused to reboot using a kernel built with the
-march=supersparc option cpuflags generated, the second bootloader
gave
'illegal access exception' after the first number appeared

[ of course I had to borrow a monitor and keyboard to see this :) ]
when I checked my kernel was 6Mb, rather than 2.1Mb , and that was after
*removing* several options....

I've now disabled the optimization on the sparc, although the PII seems
fairly happy with its use.

-- 
Rasputin :: Jack of All Trades - Master of Nuns