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.
0/x
+0
-0
x