[llvm] r319894 - [InstSimplify] Fold insertelement into undef if index is out of bounds

Igor Laevsky via llvm-commits llvm-commits at lists.llvm.org
Tue Dec 12 02:08:36 PST 2017


I reverted problematic changes in the https://reviews.llvm.org/rL320466

Is it possible to obtain an llvm-ir of the failing test case before the optimizations? I can see the miscompile, but I can’t reproduce it using optimised good case.

On 11 Dec 2017, at 19:13, Michel Dänzer via llvm-commits <llvm-commits at lists.llvm.org<mailto:llvm-commits at lists.llvm.org>> wrote:

On 2017-12-09 12:40 AM, Philip Reames wrote:
FYI, Igor is out of office for the weekend.  Feel free to revert if needed.

I guess it's possible that this change just exposed an issue in the
AMDGPU backend. Matt / Nicolai, can you take a look?

Anyway, if there's no solution available soon, it would be nice to
revert this at least temporarily until there is, because it triggered
worse issues (GPU hangs) with real-world apps, at least
https://bugs.freedesktop.org/104192 but probably more, e.g.
https://bugs.freedesktop.org/104159 .

I don't feel like the right person to make the call (or the revert).


On 12/08/2017 08:51 AM, Michel Dänzer via llvm-commits wrote:
Hi Igor,


On 2017-12-06 03:04 PM, Igor Laevsky via llvm-commits wrote:
Author: igor.laevsky
Date: Wed Dec  6 06:04:45 2017
New Revision: 319894

URL: http://llvm.org/viewvc/llvm-project?rev=319894&view=rev
Log:
[InstSimplify] Fold insertelement into undef if index is out of bounds
This change broke a bunch of piglit OpenGL tests with AMD GPUs using the
Mesa radeonsi driver. I'm attaching the LLVM IR and disassembly of the
generated code for the good and bad case of one test.




_______________________________________________
llvm-commits mailing list
llvm-commits at lists.llvm.org<mailto:llvm-commits at lists.llvm.org>
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits



--
Earthling Michel Dänzer               |               http://www.amd.com<http://www.amd.com/>
Libre software enthusiast             |             Mesa and X developer
_______________________________________________
llvm-commits mailing list
llvm-commits at lists.llvm.org<mailto:llvm-commits at lists.llvm.org>
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171212/3de34bf6/attachment.html>


More information about the llvm-commits mailing list