https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125951
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |RESOLVED
Resolution|--- |FIXED
--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Should be FIXED now.
I was tempted to look at lgrpinfo – but as it currently is only used
to determine the NUMA distance from a thread to the node that is connected
to a non-host device, i.e. a GPU used for offload, I have not added it.
(To my knowledge, only Linux and Windows are currently supported by the
vendors - and GCC narrows this further to only Linux.
Thus, it only fixes the omp_default_device omission for non-Linux code.
Thanks for the report - including pinpointing the fail in the debugger, which
helped a lot!