[Bug 102855] [OpenGL CTS] Multiple crashes in enhanced layout tests due to incorrect vue setup

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=102855

--- Comment #4 from Iago Toral <itoral at igalia.com> ---
(In reply to Iago Toral from comment #2)
> > I think we can use URB Entry Read Offset to work around this...currently we
> > hardcode it to 0 or 1...but we could probably skip over a few more.  In
> > particular, we have clip distance in the VUE slots, but the FS shouldn't
> > care, which gives us 2 more slots (1 offset)...
> 
> Yes, that's a good point... it may just be enough to make the tests pass
> although it is still not a proper fix. I'll give it a try and comment on the
> results.

I looked into it and this is what I found so far:

- gl_ClipDistance is defined as an input for fragment shaders, so we can't
blindly skip it, only if we know the FS is not reading it and only if they are
the fist builtin varying slots that come after a vue header that we are already
skipping.

- There is a lowering pass in NIR that can remap cull distances to clip
distance slots, so we can see clip distances slots in use in a FS that doesn't
read gl_ClipDistance but reads gl_CullDistance and we can't skip them in that
case (in practice this doesn't change anything for us, but I guess it is worth
mentioning)

- Even with the above caveats, the fix is enough to make the tests not crash
any more, so it is sufficient as a work around for these crashes.

I am running the patch through jenkins, if I don't see anything weird I'll send
it for review.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20170922/555c2533/attachment.html>