[adgjlsfhk1]

It's really annoying that IEEE set the exponent bias wrong. x!=0 => 1/x!=Inf was a totally achievable property if they had wanted it (by tweaking the implicit bias)