Floating-point underflow also raises FP exceptions, and it should be investigated how implementations deal with it.