atomic accesses

[Segher Boessenkool]

> This seems somewhat like other issues we have dealt with concerning 
> thread safety.  Perhaps an entire section addressing code generation 
> issues related to correct multi-threaded semantics.

I like that idea.  But, we need to decide what those correct semantics 
_are_.
Or we can wait for that to be standardised.  We should decide 
*something*
though, not just wait forever, IMHO.


Segher