Comment 7 for bug 635388

So long as the implementation of the atomics in libatomic-ops is correct, I don't see a strong need to change it. My view is that people shouldn't roll their own atomics implementations unnecessarily, but it's OK to use a library for it, so long as the library is correct.

The actual build failure here looks more like an issue with the exported headers from libatomic-ops than an issue with the _implementation_ of the operations -- is this a correct conclusion?