https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80640
Alexander Monakov <amonakov at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution|--- |FIXED
--- Comment #9 from Alexander Monakov <amonakov at gcc dot gnu.org> ---
Fixed. The fix is in the middle-end rather than the i386 backend since it was
found that at least one other target (s390) got it wrong, and fixing it once
and for all by emitting a compiler barrier in addition to whatever
"mem_thread_fence" pattern would expand to is simpler and more robust.
(corresponding issues for signal_fence and atomic loads/stores remain)
