 |
Center for Nonlinear Studies |
A novel software system, JScp, is described, which applies an AI technique called "supercompilation" to optimize programs written in the Java language. The current Java supercompiler demonstrates a powerful ability to drastically optimize a variety of different Java programs, and is actively under development aimed at improving its scope of applicability. In this talk I will first review the overall methodology of supercompilation, discuss some special aspects of the application of supercompilation to the Java language, and then discuss a few practical examples of Java supercompilation: the Fast Fourier Transform (prototypical of mathematical code, which is generally highly amenable to supercompilation), the Einstein fish puzzle (prototypical of supercompilation-as-logical-problem-solving), and a combinatory logic interpreter (prototypical of the supercompilation of language interpreters, another natural and relatively easy application domain).