An Application of Abstract Interpretation to Floating Point Arithmetic