[Bug 110432] dEQP-VK tests running very long time

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=110432

Ian Romanick <idr at freedesktop.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |NEEDINFO

--- Comment #4 from Ian Romanick <idr at freedesktop.org> ---
I think you're getting tricked.  8fb8ebfbb05 ("intel/compiler: More peephole
select") introduced a bug that caused many SPIR-V loops to not get unrolled. 
On the 19.0 branch, we fixed the bug by reverting 8fb8ebfbb05.  On master we
fixed it by b031c643491.  My guess is that those tests are fast when the loops
aren't unrolled, and they are slow when the loops are unrolled.  I bet they
will also be slow at 8fb8ebfbb05~1.

I tried a few commits with a -Og debug build and NIR_VALIDATE=0...

18.3:        128.73user 2.96system 2:12.25elapsed 99%CPU (0avgtext+0avgdata
291884maxresident)k
ca36eb12fdf: 172.17user 3.45system 2:56.55elapsed 99%CPU (0avgtext+0avgdata
290208maxresident)k
8fb8ebfbb05:   1.84user 0.13system 0:02.13elapsed 92%CPU (0avgtext+0avgdata
269244maxresident)k

I wasn't able to reproduce 38:43, but ~2 seconds -> ~2 minutes is a pretty big
jump.  What result do you get at tip of 18.3?

-- 
You are receiving this mail because:
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/20190417/c0566b31/attachment-0001.html>