fast-math-1.0: Non IEEE-754 compliant compile-time floating-point optimisations