3 ms·

And to be clear that performance win comes from removing the branches that are otherwise needed to provide x86 semantics