RE: CMM-to-ASM: Register allocation wierdness