Add relaxed atomic inc/dec intrinsics

[brson]

The exchange heap uses atomic inc/dec to track allocations. These don't require a fence. Verify that using relaxed atomics here is a win.

[brson]

Relaxed atomics generate the same assembly here.

[bblum]

Reopening this issue (at least for more discussion). I'm rewriting sync::rwlock to use an atomic uint for performance, and in some cases I'd like to use the relaxed consistency ordering.

I don't expect the assembly would change at all on a strongly-consistent architecture, unless LLVM decided on its own to reorder the instructions. I do expect that we'd see a difference on ARM -- we could save a few global barrier instructions.

[Thiez]

@bblum is correct, on x86 it will be hard or impossible to see the difference between the stronger and weaker atomics (and in some cases, even between atomics and non-atomics) because the architecture already has very strong guarantees. This presentation might be relevant:
http://channel9.msdn.com/Shows/Going+Deep/Cpp-and-Beyond-2012-Herb-Sutter-atomic-Weapons-1-of-2
http://channel9.msdn.com/Shows/Going+Deep/Cpp-and-Beyond-2012-Herb-Sutter-atomic-Weapons-2-of-2
About a third in the second video some architectures are compared.

[toddaaro]

I believe this is finished and can be closed. If I'm wrong re-open it. I'm thinking of @Aatch's set of atomic intrinsic commits.