On Fri, Apr 11, 2003 at 03:56:07PM -0700, Jason Thorpe wrote:
> 
> On Friday, April 11, 2003, at 01:43  PM, David Laight wrote:
> 
> > My Athlon 700 gains about 1.5% on 8k aligned calls.
> > For 20 byte aligned transfers the gain is 38%
> > For 20 byte misaligned transfers the gain is 29%
> 
> Can you make the same tweaks to memset()?  That is what we prefer to 
> use these days, after all :-)

bcopy/memcpy will also benefit...

Interestingly aligning bzero make almost no difference to my athlon,
however testing on a P2 (cvs.netbsd.org) shows that it makes a
considerable difference.  Anyone got a P4?

I started this looking at which copy was included in the boot images.
Several of these routines exist in more than one of libkern, libsa
and libi386.  Which means that in some cass the C routines are used :-(

	David

-- 
David Laight: david@l8s.co.uk