[Bug 100092] Assertion failed: size_log2 <= ANV_MAX_STATE_SIZE_LOG2 While compiling large compute shader

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=100092

--- Comment #2 from Jason Ekstrand <jason at jlekstrand.net> ---
Thanks for the bug report!  This is a known issue and we keep bumping the
maximum but apparently not far enough.  I'll try and look into the overflow
issue some time this week.

As far as making stuff shared goes, that's not a bad idea given that you are
already doing so for roi.  In this case, using shared memory is basically just
telling the compiler "please put this in memory, not registers" which is what
you want the compiler to do in this case.  Our compiler tries to fairly
aggressively (too aggressively in some cases) put things in registers and this
is a case where you really want it to spill.  It's interesting that you found a
perf cliff on NVIDIA as well.  Does it help perf to put depths in shared on
NVIDIA?

In any case, the fact that that array explodes your shader is also somewhat our
fault because this is definitely something our compiler could be doing for you
and I've got plans to do just that but it hasn't happened yet.

-- 
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/intel-3d-bugs/attachments/20170307/395f14f0/attachment.html>