(arith_driver, float_arith_driver): Compute (- x) by using negation, not subtraction; this makes a difference with IEEE floating point arithmetic (and also if integer arithmetic is ones' complement or signed-magnitude!).