fast-math-0.1: Non IEEE-754 compliant compile-time floating-point optimisations

Numeric.FastMath.Infinitesimal

Description

Also rewrite 0/x to +0, which should really be -0 for negative x.