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

[Hans Wennborg via llvm-commits]

https://github.com/zmodem approved this pull request.

It seems unfortunate that if someone hand-wrote the IR to make the same i128 calls, we'd still lower those incorrectly.

This still seems like an improvement though, so I'm okay with it. Maybe give Eli a day or two to respond in case he has any pointer for how to fix it properly.

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