The goal of this project is the development of compiler optimization techniques for high-level languages. The primary programming languages used today (C, Fortran and Pascal) are not very high-level. A large portion of the effort in writing software in these languages is devoted to hand-tuning the individual application to the target machine, a process that is both costly and error-prone. Higher-level languages that abstract away many of the implementation details have been developed but are not yet in widespread use. In spite of the benefits that higher-level languages promise, history suggests that they will not come into general use until programs written in them perform as well or better than those in existing low- level languages. Optimization of object-oriented programming languages, one class of high-level languages, is the major concern of this proposal. Object-oriented programming introduces two types of inefficiencies. First, many more subroutine calls are required than in normal programming methodologies. Second, these subroutines are typically more general than the code necessary to implement the function in line. Optimization of object-oriented programs primarily involves tailoring instances of subroutines to the context in which they are invoked. Automatic tailoring requires much more powerful analysis and transformation techniques than are generally available today. Developing analysis and transformation techniques to do the tailoring of these programmer-constructed abstractions is the main focus of this project.
