[clang] Bfi precision (PR #66285)

[Duncan P. N. Exon Smith via cfe-commits]

dexonsmith wrote:

Not entirely accidental. When BPI/BFI first landed it was heavily profiled to be sure it didn't pessimize non-PGO code. I don't see why we'd suddenly be okay with pessimizing it.

Under 2% isn't hard to hit for hot path code. Lots of functions will have strings of early exit conditions.

https://github.com/llvm/llvm-project/pull/66285