Intel ( i845G ) profiling

[Eric Anholt]

On Thu, 2008-03-13 at 15:25 +1100, Daniel Kasak wrote:
> On Wed, 2008-03-12 at 16:54 -0700, Eric Anholt wrote: 
> > On Wed, 2008-03-12 at 12:06 +1100, Daniel Kasak wrote:
> > > On Tue, 2008-03-11 at 11:12 -0700, Eric Anholt wrote:
> > > > > If so, the reason we're falling back is because the A8 format (0,0,0,a)
> > > > > is unsupported on that chipset.  But we've got the I8 format which
> > > > > produces (a,a,a,a), and if we're using the a8 picture as a
> > > > > non-component-alpha mask, we'll only use that fourth component.  For a
> > > > > more general implementation, those color channels could be treated as
> > > > > zero by using ARGx_SEL_ONE | ARGx_INVERT when reading them instead of
> > > > > ARGx_SEL_TEXELy.
> > > > 
> > > > Untested patch is up at:
> > > > git://people.freedesktop.org/~anholt/xf86-video-intel i845-render-fix
> > > 
> > > Cool. Thanks :)
> > > 
> > > What are the pre-requisites for this branch to build? I've built libdrm,
> > > mesa, xserver, libX11, cairo & pixman from today's git. But building the
> > > i845-render-fix branch of xf86-video-intel I get:
> > 
> > What, you want the patches I give you to actually compile? :P
> > 
> > New version pushed that's less broken, but with the same level of
> > guarantees as before (meaning: none)
> 
> Wow! That makes a big difference. EXA without a compositing manager is
> now quite fast. It's not as fast as XAA, but it's a *whole* lot better
> than before, and certainly usable. With xcompmgr it's also very good -
> this combo was previously completely unusable. With ecomorph ( compiz
> port ) it's a lot better than before ( before it was ridiculously
> slow ), but it's not _quite_ fast enough to call 'snappy'. Anyway, it's
> a huge improvement on before :)
> 
> I had to downgrade to xserver-1.4 and mesa-7.0.2 because my git build of
> these would lock hard on startup ( with any version of
> xf86-video-intel ). Oh well ...
> 
> I've tested for about 2 hours now, and haven't had any problems.
> 
> I've done another round of profiling, with expedite and our internal CRM
> package, both with & without ecomorph running. The results are at:
> http://entropy.homelinux.org/i845_render_fix_benchmarking.tar.bz2
> My original results ( with the latest released driver ) are still at:
> http://entropy.homelinux.org/intel_benchmarking.tar.bz2
> 
> Thanks again for working on this :)

OK, I pushed it with a minor correctness fix revealed by rendercheck
(took most of a day to get the 865 up and working again).  Note that the
patch you tested and that I pushed didn't replace the A8 texture type
with I8, even though A8 isn't supposed to be supported on 845 (oops). My
guess is that it just gets treated as I8 anyway.  If we see weird issues
pop up with 830/845 render, that would be the thing to check.

-- 
Eric Anholt                             anholt at FreeBSD.org
eric at anholt.net                         eric.anholt at intel.com

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 195 bytes
Desc: This is a digitally signed message part
URL: <http://lists.x.org/archives/xorg/attachments/20080313/dbd502be/attachment.pgp>