Adding optimized kernel copying support to FreeBSD-i386 - Part II

[Attilio Rao]

2006/3/6, Divacky Roman <xdivac02 at stud.fit.vutbr.cz>:
> On Mon, Mar 06, 2006 at 11:20:10AM +0100, Attilio Rao wrote:
> > Hi,
> > I did not received too much feedbacks about Part I stability, but I went
> on,
> > so Part II is available for tests.
> >
> > It presents an mmx copy which doesn't give much performance increases but
> > what I need now is testing for correctness. In particular I would see
> > feedbacks about races inside dropping/undropping FPU mechanism. It would
> be
> > better testing with PREEMPTION + FULL_PREEMPTION enabled (and maybe on SMP
> > archs).
> >
> > Some minor styling issues have been fixed of the previous patch.
> >
> > http://users.gufi.org/~rookie/works/patches/mmxcopy.tar.gz
> >
> > This mmx copy is not intended to be in final release, this is just a
> testing
> > issue in order to involve CPU < P4 & friends.
> >
> > For anything else, mail to me.
>
> hi,
>
> thnx for the great work. I think you should publish some benchmarks
> (ministate-ed buildworld before and after or something like that) and also
> tell
> people what is suitable for their processor. this way people are not
> interested
> in testing ;(
>
> anyway - thnx again for your work

Actually what I'm interested in is stability. I had a 2% measurable
clock wall improvement just for fpu overhaul (bandwidth increased 146
MB/s to 149 MB/s on my P3, 450 MHz). The main effort will be on
P4,Xeon and on amd64 porting for Athlon (I'm speaking about
performancy matters).
But now I need some very intensive testings about "base-code".

Attilio


--
Peace can only be achieved by understanding - A. Einstein