[Piglit] [PATCH] glx-oml-sync-control-timing: Make test behavior more stable

Illia Iorin illia.iorin at gmail.com
Mon Aug 6 14:17:12 UTC 2018


Hello Emil,

On Mon, Jul 30, 2018 at 2:06 PM Emil Velikov <emil.l.velikov at gmail.com>
wrote:

> HI Illia,
>
> Please don't top-post.
>
> On 30 July 2018 at 10:51, Illia Iorin <illia.iorin at gmail.com> wrote:
> > ping
> >
> > On Tue, Jul 10, 2018 at 2:25 PM Illia Iorin <illia.iorin at gmail.com>
> wrote:
> >>
> >> Hello,
> >> Full explanation can be found in
> Having a bug report is great, but the commit itself should be self
> sufficient.
>
> >> https://bugs.freedesktop.org/show_bug.cgi?id=89437#c4
> You can add this to the commit summary as
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89437
>
> >> If this does not completely clarify the issue. Please specify the
> >> question.
> >>
> >> Sincerely yours, Illia.
> >>
> >> On Tue, Jul 10, 2018 at 12:38 AM Mario Kleiner
> >> <mario.kleiner.de at gmail.com> wrote:
> >>>
> >>> On Mon, Jul 9, 2018 at 9:09 PM, Illia Iorin <illia.iorin at gmail.com>
> >>> wrote:
> >>> > The fourfold increase in the test failure condition
> >>> > is due to its unstable work. The old condition was left,
> >>> > but the test result was changed from fail to warn, added
> >>> > fail condition.
> >>> >
> >>>
> >>> Hi, can you clarify what the commit message wrt. "fourfold increase...
> >>> due to its unstable work" means? I'd like to understand the observed
> >>> issue better.
> >>>
> As you can see from git log Michel has improved the first iteration (
>
> The only problem reported as of comment 3 is an occasional failure on
> the first iteration.
> Something I believe is fixed with Michel's patch
> 375a0b8ac845e13c180dafc73cb4b8bc32fffb20
>
> This test works  unstable at some resolutions. For example, you can run
test on all resolutions  supported by your display and see unexpected fails.


> I'm struggling with your explanation in the bugzilla report. Can you
> provide a bit more detail?
>
> As I’ve understood from source of Mesa all OML sync calls are redirected
to specific implementation of windows manager.
It is not mentioned what deviation is permissible in OML Spec. While I was
running test on different configurations of gpu arch, resolution and
windows manager I came to the conclusion 200 ns is the acceptable delay
that helps us to distinguish the error in mesa from the error in the window
manager. If we do not change the error condition this part of test do not
have sense, because it fails so often and unpredictable, and this “fail”
doesn’t show incorrect work of opengl implementation.

This workaround helps to catch an error which cause in the implementation
of openGL and not in the implementation of the window manager.

Thanks,
Illia.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/piglit/attachments/20180806/cbdd96a1/attachment.html>


More information about the Piglit mailing list