When a number in a register of a floating point unit is copied to memory, it must be placed into either single precision or double precision format. This may require that the number be rounded. The default mode is to round to nearest, and to round to even in the event of a tie.