Abstract
This paper addresses the advantages and disadvantages of pre-rounding vs. post-rounding in a floating-point register. In this day when high speed computers with long registers are the norm, one may think that a rounding scheme is nearly irrelevant, because rounding cannot improve over the addition of a single bit of increased accuracy. However, the scheme is important precisely because of the speed of computation, and the extended computations that speed allows. The propagation of error then becomes a significant issue, and the scheme of rounding is the starting point for making any such subsequent analysis.