The documentation for -ffast-math says that the compiler can assume that no floating-point values are NaNs. Thus we can already assume that compares do not trap when -ffast-math is used. If you want to say that other operations, like divide, don't trap, then we would need another option for it Jim