https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119892
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |burnus at gcc dot gnu.org
--- Comment #1 from Tobias Burnus <burnus at gcc dot gnu.org> ---
I also see on gfx1100 for an offloading build two fails which look related
(done on April 14, 2025). However, those already failed with an older ROCm than
6.4.0, I guess the it was the mentioned 6.3.4.
(That's x86-64 + GCN offload; the results in comment 0 are for using gcn as
embedded system, running the 'normal' testsuite on the GPU.)
Namely,
* libgomp.fortran/nestedfn5.f90 - fails with -O1 or higher, works with -O0
* libgomp.fortran/pr66199-6.f90 - fails with -O3, but works with -O2
The failure is:
libgomp: GCN fatal error: Asynchronous queue error
Runtime message: HSA_STATUS_ERROR_INVALID_ISA: The instruction set architecture
is invalid.
* * *
I wonder how recent the PASS of comment 0 were - or in other words: Is this
really a ROCm 6.3.4 -> 6.4.0 fail - or could it be due to other changes done
on/before April 14 that caused my issue and Thomas issue in comment 0?
