[Mesa-dev] [Bug 94955] Uninitialized variables leads to random segfaults (valgrind log, apitrace attached)

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=94955

--- Comment #22 from Roland Scheidegger <sroland at vmware.com> ---
(In reply to David Lonie from comment #21)
> > I wonder what the test expects, I haven't seen anything but NaN coords with
> > that sampler ;-). That can't have possibly given really useful results, unless
> > not crashing is good enough to pass :-).
> 
> The test is not passing, but it is rendering a portion of the scene (the
> part that doesn't use floating point textures).
Ok. I guess the all nans there as coords have something to do with it...

> 
> Illegal sampler view creation without bind flag
This is "typically" harmless for now at least (especially on sw renderers), but
an abuse of bind flags (though I don't know about swr, if you make use of the
flags to track potential changes of the contents of resources). If a resource
is created, it must have sampler view bind flag, otherwise it's illegal to
later create a sampler view. The problem is, this requirement doesn't fit well
with GL. Sometimes for things as blitting such resources also util code may
create sampler views out of them. If you're interested it should be pretty easy
to figure out where the result was created, why it doesn't have the sampler
view bind flag and why later a sampler view is created regardless, but it most
likely isn't the source of the issue.

> Should this be a new bug? Is there more information I can provide? This can
> be reproduced in VTK as described in comment #9.
Yes this should be a new bug (even the ones fixed actually have nothing to do
with the bug title already, so let's not make this worse).
You probably need to track down what's going wrong. All I can say from when I
was looking at the NaN coords, that shader seemed to source the coords directly
from an input attribute, and the vs looked fairly simple, so I don't know what
happened really. However, the test seemed to possibly use some kind of
multi-pass operation, hence it's possible previous shaders didn't output the
right result already.

-- 
You are receiving this mail because:
You are the assignee for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20160426/16f6f654/attachment-0001.html>