[llvm] [AMDGPU] Extend wave reduce intrinsics for i32 type (PR #126469)

[Joseph Huber via llvm-commits]

jhuber6 wrote:

> > > We need clang builtins and unit tests to ensure these really work
> > 
> > 
> > I could put some unit tests in `libc` if needed. I currently test the `scan` helpers in `gpuintrin.h` there.
> 
> Are we looking for tests to check the codegen only or do we need execution tests?

I have runtime tests in https://github.com/llvm/llvm-project/blob/main/libc/test/integration/src/__support/GPU/scan_reduce.cpp#L74, but I'm unsure if it's a fit for this since it'd need to work for NVIDIA as well. I suppose those will get tested automatically once I change the definitions in `gpuintrin.h` to use these builtins for AMD.

https://github.com/llvm/llvm-project/pull/126469