labrinea wrote: @jroelofs, aside from semantics which I'll clarify with Andrew, I am thinking that the constant folding part might be redundant and whether `FeatureBitset` is an alternative to the TargetParser. https://github.com/llvm/llvm-project/pull/87939