Comment by jimmies

If your input is Russian mixed with English, it might still be actually faster to do a first pass to process all the ASCII English range first with the aforementioned techniques, then do Russian later in the second pass: imagine you’ll have way less if branches to do.