[virglrenderer-devel] openarena workaround

[Marc-André Lureau]

Hi Dave

On Wed, Jun 29, 2016 at 11:00 AM Dave Airlie <airlied at gmail.com> wrote:

> Hey Rob (and list),
>
> I finally found some time to look at the OA performance under PTS, and
> tracked it down to the demo level PTS uses having some flares in it
> and the OA flare code doing 8 individual 1x1 readpixels on the depth
> buffer. This is clearly not a great idea, and with virgl is an extra
> bad idea.
>
> I've created a mesa branch:
> https://cgit.freedesktop.org/~airlied/mesa/log/?h=virgl-readpixels-sync
>
> That attempts to fix this, the branch does better tracking of resource
> reading/writing, and then when it sees a transfer for a 2D resource
> that is out of sync with the host copy, it does a complete surface
> sync and then doesn't readback from the host for any subsequent
> transfers until the surface is dirtied again.
>
> I haven't ran this through a complete piglit run and I'm not sure it
> won't destroy some other apps speed, but it would be good if anyone
> has time to give it a spin with some other apps/games.
>

When running piglit, I noticed half the tests failed, crashing. I fixed
your patches (correcting offset, and fixing vtest),I ran piglit:

with: [52799/52799] skip: 34638, pass: 17661, warn: 2, fail: 465, crash: 33
without: [52799/52799] skip: 34638, pass: 17646, warn: 2, fail: 479, crash:
34

Also did a naive measurement on the piglit run time takes ~25% less with
your patch.

My fixed branch is:
https://github.com/elmarco/mesa airlied



Dave.
> _______________________________________________
> virglrenderer-devel mailing list
> virglrenderer-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/virglrenderer-devel
>
-- 
Marc-André Lureau
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/virglrenderer-devel/attachments/20160806/b68e2aa6/attachment.html>