[PATCH] D11382: x86 atomic: optimize a.store(reg op a.load(acquire), release)

[JF Bastien]

jfb added a comment.

In http://reviews.llvm.org/D11382#210671, @morisset wrote:

> LGTM for the integer part.
>
> I am not really convinced that the floating point change belongs in the same patch: it is conceptually different, and does not seem to share any code with the rest of the patch. It is also not obvious to me why it is only tested on X86_64, a comment would be appreciated.


I improved the comment for x86-32. LLVM's code generation was silly when I was testing it out. The instructions I use are in SSE, and even with `-mattr=+sse` (and `+sse2`) it wasn't particularly good code, even without using any atomics. I figure x86-32 optimizations of floating-point atomics isn't particularly useful compared to x86-64 at this point in time.


http://reviews.llvm.org/D11382