https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124362
Bug ID: 124362
Summary: [16 Regression][OpenMP with GCN offload] Wrong-code
since r16-7825-gf23a339a686ed6 (Backpropagate more
equivalences in DOM)
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: openmp, wrong-code
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: burnus at gcc dot gnu.org
CC: ams at gcc dot gnu.org, law at gcc dot gnu.org
Target Milestone: ---
Target: gcn
This is about an OpenMP with offloading to an AMD GCN GPU (here: gfx90a/MI210)
and using the test progream omptests' t-dpf,
https://github.com/doru1004/omptests/blob/main/t-dpf/test.cpp
Before commit r16-7825-gf23a339a686ed6, the testcase is successful.
After the commit, running the program fails with:
$ .../g++ -fopenmp -O3 t-dpf/test.cpp -foffload=amdgcn-amdhsa
$ LD_LIBRARY_PATH=... ./a.out
no schedule clauses
GCN Kernel Aborted
The change between successful (17× "Succeeded") → 'GCN Kernel Aborted' seems to
happen since
commit f23a339a686ed6cc6a4838459bc220e48ba901cb
Author: Jeff Law
Date: Sat Feb 28 08:54:23 2026 -0700
[PR tree-optimization/90036] Backpropagate more equivalences in DOM
...
PR tree-optimization/90036
gcc/
* tree-ssa-dom.cc (back_propagate_equivalences): Accept new
argument for available expression stack. Lookup equivalences
in the expression hash table too. If an expression hash a
known constant value, record it and recurse.
(record_temporary_equivalences): Back propagate for a subset of
edge equivalences.
* * *
Disclaimer: The problem seems to be some kind of race or uninitialized variable
or ... because when it fails varies.
- In the output above, it failed right away for the first subtest.
- Running it again, it was successful for the first six tests – only failed for
the 7th test, namely:
no schedule clauses
Succeeded
schedule static no chunk
Succeeded
schedule static chunk
Succeeded
schedule dynamic no chunk
Succeeded
schedule dynamic chunk
Succeeded
dist_schedule static no chunk
Succeeded
dist_schedule static chunk
GCN Kernel Aborted
team master not responding; slave thread abortingteam master not responding;
slave thread abortingteam master not responding; slave thread abortingteam
master not responding; slave thread abortingteam master not responding; slave
thread abortingteam master not responding; slave thread abortingteam master not
responding; slave thread abortingteam master not responding; slave thread
abortingteam master not responding; slave thread abortingteam master not
responding; slave thread abortingteam master not responding; slave thread
abortingteam master not responding; slave thread abortingteam master not
responding; slave thread abortingteam master not responding; slave thread
abortingGCN Kernel Aborted
GCN Kernel Aborted
GCN Kernel Aborted
GCN Kernel Aborted
- Trying it again, it was successful for 16 subtests and only failed for the
17th.
- And running it again: 3× successful and then the 'GCN Kernel Aborted'.
* * *
This nature of the problem makes it harder to reduce the testcase. I tried
commenting out some chunks in the code and running it in the loop – but that
often made the whole program pass. :-/
