I looked briefly at optimizing InterpreterMacroAssembler::increment_mask_and_jump a bit, but it looks that current code is the best we can do. This improvement does a few related cleanups without semantic changes.