On Mon, Jun 10, 2002 at 02:33:30AM -0400, Geoff Adams wrote:
> On Saturday, June 8, 2002, at 09:21 PM, Charles Shannon Hendrix wrote:
> 
> >So before going further, I'd like to know of any success stories, and
> >ideas on why I cannot generate a good kernel, tuned for the supersparc
> >chips.
> >
> >The builds have been attempted on a pair of SS5's, one 110MHz microSPARC
> >and the other a 170MHz turboSPARC.
> 
> Hmmmm. This isn't what you're asking, but since neither of those is a 
> supersparc I wonder if you might get better performance with plain -mv8 
> (which, as you know, theoretically won't reorder the instructions to 
> suit a supersparc). I understand that you need it to work, first, and 
> only then can you compare. :)

According to the consensus in multiple discussions of this matter,
the turbo and micro do respond well to -mcpu=supersparc.  

Several people have said that they are also part of the supersparc family.

Sun's web site long ago lost coherence to me, so I've not really looked
there for an official answer, and I doubt that will tell me how gcc
options affect their CPUs anyway... :)

When I did some quick testing with openssl, I got better performance
using -mcpu=supersparc rather than -mv8 alone.  But there was nothing
scientific about my tests since -mv8 was such a drastic benefit alone.

In any case, it seems the problem is the toolchain and upgrading needs to
happen.

Cross-compiling on my faster workstation sounds interesting.  I really
hate to upgrade a stable 1.5.2 server pair, but maybe 1.6 will be
compelling enough to make it happen.

-- 
UNIX/Perl/C/Pizza__________________________________shannon@widomaker.com