[llvm] [AMDGPU] Implement IR variant of isFMAFasterThanFMulAndFAdd (PR #121465)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I'll update the test once the pre-commit PR lands.

Should use stacked PRs instead of manually referring to dependent PRs 

https://github.com/llvm/llvm-project/pull/121465