[llvm-bugs] [Bug 35983] New: piglit builtin-float-cos-1.0 crashes with llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:916: llvm::SelectionDAG::~SelectionDAG(): Assertion `!UpdateListeners && "Dangling registered DAGUpdateListeners"' failed.

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=35983

            Bug ID: 35983
           Summary: piglit builtin-float-cos-1.0 crashes with
                    llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:916:
                    llvm::SelectionDAG::~SelectionDAG(): Assertion
                    `!UpdateListeners && "Dangling registered
                    DAGUpdateListeners"' failed.
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Keywords: regression
          Severity: enhancement
          Priority: P
         Component: Backend: AMDGPU
          Assignee: unassignedbugs at nondot.org
          Reporter: pavel.ondracka at email.cz
                CC: jan.vesely at rutgers.edu, llvm-bugs at lists.llvm.org,
                    Matthew.Arsenault at amd.com

Created attachment 19693
  --> https://bugs.llvm.org/attachment.cgi?id=19693&action=edit
compressed full backtrace

Three piglit tests:
builtin-float-cos-1.0.generated
builtin-float-sin-1.0.generated
builtin-float-tan-1.0.generated
crash with llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp:917:
llvm::SelectionDAG::~SelectionDAG(): Assertion `!UpdateListeners && "Dangling
registered DAGUpdateListeners"' failed.

This is a regressing. I have run a regression tests, however due to unrelated
build failure I ended with quite lot of skips. By manually reverting the
related looking ones on top of first bad building revision (a83daa1) I found
that two commits must be reverted to get the test working again. The first one
is the likely culprit, while the second one is probably only a build fix. 

commit f68b9beeb97bd44cc651d519c45e8ad32067825d
Author: Jan Vesely <jan.vesely at rutgers.edu>
Date:   Mon Dec 4 23:07:28 2017 +0000

    AMDGPU/EG: Add a new FeatureFMA and use it to selectively enable FMA
instruction

    Only used by pre-GCN targets
    v2: fix predicate setting for FMA_Common

commit a7574026a1ac25d5955e8e29c4b5ec59ac0f6576
Author: Matt Arsenault <Matthew.Arsenault at amd.com>
Date:   Tue Dec 5 03:15:44 2017 +0000

    AMDGPU: Fix missing subtarget feature initializer

The testing was done with latest mesa, libclc and CAICOS GPU.

Bug 35910 may be related. It has the same symptoms, however as the app
mentioned there was not working even before the aforementioned commits, it is
unclear if this is the same root cause.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180117/5dfc7dd3/attachment.html>