EXA - current status with Intel driver

Clemens Eisserer linuxhippy at gmail.com
Sun Sep 23 09:55:11 PDT 2007


Hi,

I wonder when all the glucose-stuff will be useable, does anybody know
further details? I also tested EXA (although not the current work, so
I can't comment on the state-of-the-art) and was ... well ... not so
happy as it actually made many things worse (although even the
XAA-driver seems to heavily rely on libfb-software routines).
It may improve many operations and open the door to accalerate things
that are now doomed to software, but overall responsivness suffered
:-/

Does glucose work with the intel-drivers, and how hard would it be for
closed-source driver programmers to make their drivers glucose
compatible?

Thanks, lg Clemens

2007/9/23, Lukas Hejtmanek <xhejtman at ics.muni.cz>:
> On Thu, Sep 13, 2007 at 06:57:57PM +0200, Michel Dänzer wrote:
> > > > Ian Romanick did most of the pci-rework changes, though this might also
> > > > be related to the corresponding changes in the intel driver. FWIW though
> > > > I tried the master branch briefly on my i945 machine before pushing and
> > > > didn't seem to see this problem.
> > >
> > > So with the current head, you are able to get 1400+FPS in x11perf -shmput500?
> >
> > I think so, but I'm away from that machine and probably won't be able to
> > confirm for another week at least.
>
> This problem seems to be gone with latest head of all the stuff (Xserver,
> drivers, mesa, drm, etc).
>
> However, overal desktop experience is still worse than with XAA. Your EXA
> improvements seems to help a lot compared to the previous state. However,
> firefox scrolling eats so much CPU that ondemand governor switches to high
> frequency, this is not the case with XAA.
> (this page is taken as reference:
> http://gitweb.freedesktop.org/?p=xorg/driver/xf86-video-intel.git;a=summary)
>
> Text scrolling in gnome-terminal is much slower and much more CPU consuming.
>
> Last but not least, resume from RAM does not work. (Not sure whether EXA or
> new Eric's stuff with buffer objects is the reason.)
>
> --
> Lukáš Hejtmánek
> _______________________________________________
> xorg mailing list
> xorg at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/xorg
>


More information about the xorg mailing list