implicit normalization of floating point numbers