> Why not do that round-robin as the comment suggest? I agree, round robin register allocation would give the biggest gains followed by improving dependency analysis across extended basic blocks. Stan