[Piglit] [PATCH 0/5] Randomized UBO tests of doom

Ian Romanick idr at freedesktop.org
Wed Jan 13 15:05:27 PST 2016


On 01/13/2016 01:34 PM, Dylan Baker wrote:
> I seem to remember this, it was nvidia I think, if that helps you look
> for the email.

Looking at old messages in the thread... it looks like this was fixed in
shader-runner in
http://cgit.freedesktop.org/piglit/commit/?id=2b94faec18dc1f8f0d9241ec731408959320cd7c.

> On Wed, Jan 13, 2016 at 11:39 AM, Ian Romanick <idr at freedesktop.org
> <mailto:idr at freedesktop.org>> wrote:
> 
>     There is at least one bigger bug in this series, and fixing it will
>     require changes to shader-runner too.  I had some e-mail either from one
>     of the guys at VMware or one of the guys at NVIDIA, but I can't find it.
> 
>     My recollection is that there is some case in the generator (and
>     shader-runner in general) that expects non-base elements of arrays to be
>     advertised, but the spec says they should not be advertised.  I think
>     the case was that for a UBO like
> 
>     struct S {
>         int i;
>         float f;
>     };
> 
>     uniform U {
>         S s[4];
>     };
> 
>     only s[0].i and s[0].f would be reported, but shader-runner expects all
>     elements of s[] to be reported.
> 
>     This will be annoying to fix because shader-runner relies on this to
>     determine the offsets s[1].i and friends.
> 
>     It should be relatively easy to figure out what the failure case is.  I
>     think there's already a test in piglit that misuses the feature, and
>     that test fails on NVIDIA's closed-source driver.
> 
>     On 01/12/2016 03:09 PM, Ilia Mirkin wrote:
>     > My fixes are one-liners (might just be one, I forget). At this point I
>     > would very much like to see *anything* checked in, and we can improve
>     > from there.
>     >
>     > On Tue, Jan 12, 2016 at 6:04 PM, Dylan Baker
>     <baker.dylan.c at gmail.com <mailto:baker.dylan.c at gmail.com>> wrote:
>     >> That sounds like a plan. Would you like to get Ian's original scripts
>     >> merged, and I'll update the cleanups I have for his scripts and
>     send them
>     >> out?
>     >>
>     >> Dylan
>     >>
>     >> On Tue, Jan 12, 2016 at 2:41 PM, Timothy Arceri
>     <t_arceri at yahoo.com.au <mailto:t_arceri at yahoo.com.au>>
>     >> wrote:
>     >>>
>     >>> On Tue, 2015-11-10 at 16:46 -0800, Dylan Baker wrote:
>     >>>> Since gravy is so delicious,
>     >>>>
>     >>>> https://github.com/dcbaker/piglit.git wip/ubo-fuzzer
>     >>>>
>     >>>> bin/piglit run ubo-fuzzer output
>     >>>>
>     >>>> Obviously it's not at all feature complete, it's more proof of
>     >>>> concept
>     >>>> than anything else, but gravy.
>     >>>
>     >>> Hi Dylan,
>     >>>
>     >>> I took a quick look at your branch and I'd like to suggest a way
>     >>> forward on this.
>     >>>
>     >>> It looks like you made a bunch of small/fixes tidy ups to Ian's
>     >>> scripts, how about we get those sent out for review/comment then
>     commit
>     >>> the script to the main piglit repo.
>     >>>
>     >>> Ilia and I can then rebase our modifications, resend for review and
>     >>> commit those.
>     >>>
>     >>> Then we can worry about gravy once we have all the work in one
>     place.
>     >>>
>     >>> How does that sound?
>     >>>
>     >>> Tim
>     >>>
>     >>>> On Tue, Nov 10, 2015 at 03:01 11AM -0500, Ilia Mirkin wrote:
>     >>>>> That's a full step ahead of my suggestion. I just want the
>     thing in
>     >>>>> a
>     >>>>> shared place so that I can run it when doing dodgy things to
>     code I
>     >>>>> don't understand... if there's some automated process running it,
>     >>>>> that's just gravy on top.
>     >>>>>
>     >>>> [snip]
>     >>
>     > _______________________________________________
>     > Piglit mailing list
>     > Piglit at lists.freedesktop.org <mailto:Piglit at lists.freedesktop.org>
>     > http://lists.freedesktop.org/mailman/listinfo/piglit
> 
> 



More information about the Piglit mailing list