On the supremum in fractional programming