https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120530
Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |ASSIGNED
Target Milestone|--- |15.2
Ever confirmed|0 |1
Last reconfirmed| |2025-06-05
Assignee|unassigned at gcc dot gnu.org |burnus at gcc dot
gnu.org
Keywords| |patch
--- Comment #3 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
(In reply to Tobias Burnus from comment #2)
> > That system has unified shared memory
Ah...
> Indeed, if I compare the host and the device addresses, I see
> different values (= mapped) for sptr1, sptr1->ptrset, sptr1->ptrset2 (→ OK)
>
> BUT: sptr1->ptrset[i] and sptr1->ptrset2[i] still have the host value!
>
> Can you confirm that it works with the following patch?
Yes, that makes it PASS for all the several nvptx offloading configurations
I've tried.
Do we (that is, you) understand why this wasn't a problem for GCN offloading,
without USM?
