<<For floating point: 0.0/0.0 might generate a floating point exceptions and the result is a NaN. The same applies for the infinites. >> Of course this is only required if you are trying to be strict IEEE, I would think it perfectly fine to do this optimization unless -mieee were set.