I\'m facing the same situation for several times now: I have an approximation of a given function (e.g. cosine) in single precision, and I want evaluate the maximum error of my
