Re: about powerpc version of in{,4}_cksum

[Matt Thomas <matt%3am-software.com@localhost>]

At 12:58 PM 8/2/2002, Ignatios Souvatzis wrote:
> Regards,
>
> On Wed, Jul 31, 2002 at 08:44:12PM -0700, Matt Thomas wrote:
> > At 11:07 AM 7/30/2002, Matt Thomas wrote:
> > >Well, for the G4 case I plan on getting a AltiVec implemention
> > >of in_cksum working shortly.
> >
> > Well, I've written a such a beast and it tends to be 3 times
> > as fast as either the netinet or the powerpc version.  The cross
> > over point where AltiVec is faster seems to be about 32 bytes.
> >
> > I have a few more things to clean up but it's looking very
> > promising.  I hope to check it in in a few days.
>
> Wouldn't this force us to save the Altivec registers not only on user-user
> context switch, but already at kernel entry?

No.  Just like the pmap AltiVec routines, the in_cksum routine will save
and restore any registers it uses so that the lazy AltiVec switching
will continue to work properly.


--
Matt Thomas               Internet:   matt%3am-software.com@localhost
3am Software Foundry      WWW URL:    http://www.3am-software.com/bio/matt/
Cupertino, CA             Disclaimer: I avow all knowledge of this message