The softfloat implementation is still missing (only corner cases are supported). Extend softfloat floatNN_class methods to distinguish between SNaN and QNaN.