The impact of the proposed standard for floating point arithmetic on languages and systems

Abstract

Most programming languages seem to take a grudging approach to floating point computation. Many processors support these operations, and some peculiar and important users insist on using them, so languages provide some facility for using floating point entities. But very little care seems to be taken with numeric issues. Part of the problem is the wide range of anomalous hardware that must be made to "fit" the language. A language definition can be very vague (the most popular approach), or it can cover the limitations and anomalies by a careful specification (as in Brown's model [1]), or it can restrict its domain to particular hardware (such as the Floating Point Standard). The main impact of the proposed floating point standards may be to force a reconsideration of the role of floating point in programming languages. The following comments apply to all of the major proposals; where details are needed for a specific example, the Kahan, Coonen, and Stone [2] proposal is used.