[llvm] [PowerPC] improve performance on the isNan and !isNan function in case of -ffp-model=strict" (PR #215587)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> > I don't think there's any point in doing this. This IR is not canonical. instcombine turns all of your test functions into fcmp ord/uno as-is
> 
> I implement customer lower in the PPCTargetLowering::LowerIS_FPCLASS

This still doesn't address that this is optimizing non-canonical IR. The backend should only be optimizing patterns that can appear out of legalization. Cases like this should be constructed in the canonical form to start

https://github.com/llvm/llvm-project/pull/215587