We prefer to use size and shift than to get the size by reading the mask
because the bits in mask might not be contiguous.