[Piglit] [PATCH 00/29] Dispatch support for EGL and GLES

[Chad Versace]

Pauli, I plan to look at these patches tomorrow. Below are a few comments
unrelated to the patch series.

On 05/21/2012 11:08 AM, Pauli Nieminen wrote:

> The patches are "waffle ready" meaning that there is some differences what is done if piglit is compiled using waffle to support runtime selection of EGL/GLX and GL api.
> 
> I did run quick tests set on i965 to check for regressions. To bad it appears
> that piglit can hit random failures before and after. I tried to reproduce
> random failures manually but failed so far.
> 
> I already saw once random failure when I was running bisectability tests on
> small set of tests. In that case it appeared like some rendering didn't hit the
> target memory before glReadPixels. But I couldn't reproduce the issues with
> while <test>; do sleep 0; done.

I've observed that the rate of inconsitencies increase when using a compositing
window manager. This is known, long-standing issue that we don't know to diagnose.
I recall that Anholt had some theories about the root cause that involved Mesa and
xf86-video-intel incorrectly sharing gem buffers.
 
> Uploaded results are in:
> http://people.freedesktop.org/~suokko/quick/changes.html
> 
> The branch ontop of piglit/master:
> http://cgit.freedesktop.org/~suokko/piglit/log/?h=egl_and_gles_dispatch
> 
> The branch on top of Chad's waffle-v3.
> http://cgit.freedesktop.org/~suokko/piglit/log/?h=waffle_egl_and_gles_dispatch

FYI, I've pushed waffle-v8, which builds against waffle-0.3.
On that branch, all tests succesfully build and there is only a single
test regression. I'm almost ready to submit those patches.
 
----
Chad Versace
chad.versace at linux.intel.com