[llvm] [X86] Add bf16 support to isFMAFasterThanFMulAndFAdd for basic FMA optimizations (PR #172006)

[Phoebe Wang via llvm-commits]

================

----------------
phoebewang wrote:

I wonder if we can lower to F32 FMA instruction if we allow for `hasBF16() || hasAVXNECONVERT()`.

https://github.com/llvm/llvm-project/pull/172006