Floating-Point Rounding Errors: Absolute and Relative Errors