Steps toward an APL compiler

Abstract

Compilation of APL programs could yield a dramatic improvement in performance for some kinds of programs, provided that the compiler could simplify the processing to the actual requirements of the particular algoritlam.Other optimization approaches have been unable to achieve the performance improvements that seem to be possible. Compilation aided by declarations has been demonstrated, but it is undesirable to require declarations if most of the information they contain can be deduced by inspection of the program.Certain changes and restrictions to the language would make such deduction feasible by simplifying the analysis of control flow and removing apparent interfaces between programs that are not actually used as interfaces.Some of the proposed language changes are regarded as improvements because they facilitate deductions about programs by humans and therefore contribute to program readability and reliablility.The other restrictions are inconvenient when debugging a program, but the impact is minimal because compilation of APL would be appropriate mainly for production programs that already have been tested. PERFORMANCE DEFICIENCIES OF PRESENT APL SYSTEMSThe performance of current APL systems is not acceptable for some applications.Of course, the same can be said of any high level language, since it is always possible to write more efficient programs in assembly language.

An algebraic array shape inference system for MATLAB®

“…At the source-level, the execution of an array expression could be thought of as a two-phase process: a preprocessing phase in which it is examined and prepared for evaluation, and an evaluation phase in which all the elements of its array result are computed [Wiedmann 1979]. The main characteristic of the preprocessing phase is that it is composed of subtasks that can be generally performed in a time that is independent of the number of elements in the array arguments to the • P. G. Joisha and P. Banerjee expression.…”