The elimination of intermediate containers in the evaluation of first-class array expressions

Abstract

A language with first-class arrays has functions that map from arrays to arrays, where arrays are nested rectangular containers. Such languages are conceptually appropriate for a wide range of applications. Unfortunately, straightforward implementations have high memory allocation and data movement overheads. An algorithm and implementation which discover efficient implementations for first-class array expressions is presented. This optimizer achieves results equivalent to the code a good imperative programmer would produce. For many examples, all temporary arrays and unnecessary moves are avoided. To motivate and provide a basis for this work, a first-class array language is defined and examples of its use are given.<>