[Bug libstdc++/54005] Use __atomic_always_lock_free in libstdc++ is_lock_free instead of __atomic_is_lock_free

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54005

--- Comment #34 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
(In reply to Jonathan Wakely from comment #32)
> I would prefer a function template instead of the LF macro e.g.

That will lose the __LINE__ information from the supposed error message in
VERIFY, i.e. it'll point at the check function rather than the unique invoking
line.

> So the premise that "is_lock_free() is per-type implies it's the same as
> always_lock_free" is wrong.

Ouch.  That again exposes the __is_lock_free() flaws.  But that's for another
bug.

Can I at least change 
-           reinterpret_cast<void *>(-__alignof(_M_i)));
+           reinterpret_cast<void *>(-_S_alignment));
?