Adder errors versus multiplier errors in floating-point digital filters: