explicit normalization of floating point numbers