SELECTIVE SPECIALIZATION FOR OBJECT-ORIENTED LANGUAGES

Dynamic dispatching is a major source of run-time overhead in object-o riented languages, due both to the direct cost of method lookup and to the indirect effect of preventing other optimizations. To reduce this overhead, optimizing compilers for object-oriented languages analyze the classes of objects stored in program variables, with the goal of b ounding the possible classes of message receivers enough so that the c ompiler can uniquely determine the target of a message send at compile time and replace the message send with a direct procedure call. Speci alization is one important technique for improving the precision of th is static class information: by compiling multiple versions of a metho d, each applicable to a subset of the possible argument classes of the method, more precise static information about the classes of the meth od's arguments is obtained. Previous specialization strategies have no t been selective about where this technique is applied, and therefore tended to significantly increase compile time and code space usage, pa rticularly for large applications. In this paper, we present a more ge neral framework for specialization in object-oriented languages and de scribe a goal-directed specialization algorithm that makes selective d ecisions to apply specialization to those cases where;it provides the highest benefit. Our results show that our algorithm improves the perf ormance of a group of sizeable programs by 65% to 275% while increasin g compiled code space requirements by only 4% to 10%. Moreover, when c ompared to the previous state-of-the-art specialization scheme, our al gorithm improves performance by 11% to 67% while simultaneously reduci ng code space requirements by 65% to 73%.