No idea if it's not already optimised, but x2 could also be x*x and not just abs_x * abs_x, shifting the dependencies earlier.