A comparison of two recent approaches to machine parameterization for mathematical software
Brian T. Smith · ACM SIGNUM Newsletter · 1978
Recently, there have been published two different proposals for expressing the dependence of numerical software on the environment. One proposal, which is essentially described in two papers [1,2]** characterizes the dependence of the software on the machine architecture essentially in terms of the representation of floating point entities in the machine. The second, more recent proposal, described in [3] characterizes the dependence of the software on the environment in terms of models both the floating point numbers and of the behavior of the arithmetic unit of the machine. The purpose of this paper is to expose and clarify the differences and similarities between these two approaches. It is the author's opinion that the two approaches interpreted in their strictest sense serve two distinct purposes. When each is interpreted imprecisely, their roles overlap. In order that the comparisons briefly discussed below are fully understood, we need to briefly summarize the separate approaches. Since our interest here is in the parameterization of the numerical aspects of the environment, the non-numerical parameters discussed in [1] and the considerations of implementation discussed in [2,3] will not be considered here.