[llvm] [InstCombine] Canonicalize reassoc contract fmuladd to fmul + fadd (PR #90434)

[Andy Kaylor via llvm-commits]

andykaylor wrote:

I'm curious how you arrived at this case. I wouldn't expect clang to generate the llvm.fmuladd intrinsic in the fp-contract=fast state. If another front end is doing it, I wonder if they intended semantics other than what they will actually get. here. Or perhaps some transformation incorrectly set the 'contract' flag.

https://github.com/llvm/llvm-project/pull/90434