[llvm] [X86] Correctly call 16 byte atomic helpers on x86_64 Windows (PR #181356)

[Martin Storsjö via llvm-commits]

mstorsjo wrote:

Now this feels rather hackish and ugly, to work around it like this. But this is how it works for code generated by Clang anyway, and I don't know the LLVM lowering internals well enough to attack the task of fixing the lowering of i128 arguments/return values correctly.

https://github.com/llvm/llvm-project/pull/181356