[Bug tree-optimization/54003] New: __atomic_always_lock_free inconsistent with __atomic_is_lock_free

[hp at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54003

             Bug #: 54003
           Summary: __atomic_always_lock_free inconsistent with
                    __atomic_is_lock_free
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: hp@gcc.gnu.org
                CC: amacleod@redhat.com


Created attachment 27817
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27817
Testcase suitable for gcc.dg/torture

The attached code, suitable for gcc.dg/torture (or add -O2 to options if you
want it to play in the same pool as the other atomic tests), exposes an
inconsistency in the atomic support.  The "constant" test says true while the
"varying" test is allowed to say (and forced to say) false.  I think this
particular case is curable by the to-be-attached patch, but I haven't tested it
and the comment needs tweaking, so just passing along as conceptual. :)