[llvm] AMDGPU: Optimize set_rounding if input is known to fit in 2 bits (PR #88588)

[Joseph Huber via llvm-commits]

https://github.com/jhuber6 approved this pull request.

Seems pretty straightforward given we already have `llvm.get_rounding` implemented.

https://github.com/llvm/llvm-project/pull/88588