Masters Theses

Date of Award

8-1999

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

Bruce Whitehead

Committee Members

D. P. Mehta, Al Pujol

Abstract

Dynamic binding (dispatch) in object-oriented languages prevents compilers from performing static optimizations, such as method inlining. Type Inference, a static optimization technique, helps eliminate dynamic binding by providing compilers with receiver class type information for virtual method invocations during compile time instead of runtime. Type Inference has been implemented on SELF, a pure object-oriented language, and proven effective by inlining (statically binding) 95% of virtual method calls. It has also been tested on other object-oriented languages, and results indicate that the technique is just as effective on other languages as it is on SELF. This research evaluates the effectiveness of Type Inference on Java Bytecodes using the Cartesian Product Algorithm. The technique is applied to a bytecode-to-bytecode optimizer, which reads and parses a Java class file, perforins dataflow analysis and type inference on the bytecodes, performs optimization by eliminating virtual method invocations, and recompiles the modified bytecode into a new Java class file. Evidence resulting from this research indicates that Concrete Type Inference is an effective optimization technique for object-oriented languages, including Java. More importantly, the technique was successfully implemented and tested on Java Bytecodes. The results are not outstanding; however, some improvements are evident from the results. Further improvements should be possible by implementing other optimization techniques, such as method inlining and side-effect analysis.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS