[Gc] Should libatomic_ops be inside bdwgc?

[Petter Urkedal]

On 2011-08-17, Ivan Maidanski wrote:
> Hi Petter and Hans,
> 
> Is the current libatomic_ops API sufficient to implement the complete C++0x atomics API? (Or, we don't need to implement all the C++0x primitives, do we?)

I can answer part of that question.  The AO_t may be to narrow to
implement all the atomic_X_t types.  In the ISO/IEC:TC2 draft which I'm
using as my C99 reference, the corresponding stdint.h are optional
(7.18.1.1.3):

    These types are optional. However, if an implementation provides
    integer types with widths of 8, 16, 32, or 64 bits, no padding bits,
    and (for the signed types) that have a two’s complement
    representation, it shall define the corresponding typedef names.

Not sure if that changed in the final standard.  On the other hand N1349
does not state that the atomic_X_t types are optional, but since my
reference[1] is a proposal, that may change.

In any case, I think it'll suffice for our purpose to only implement up
to the word size.  In the long run the stdatomic.h header will be
shipped by the compiler vendor anyway.

> A sub-question: what's about CAS? I mean do we need a CAS variant returning old-value to implement the corresponding C++0x primitive?

[1] http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1349.htm