[clang] [AMDGPU] Use generic builtins for `wave_reduce` ops (PR #179589)

[Alex Voicu via cfe-commits]

AlexVlx wrote:

> This also breaks using __has_builtin for a specific type + target

I don't quite understand what you mean, do you have an actual example.

https://github.com/llvm/llvm-project/pull/179589