While extensive research efforts have been expended on developing compiler optimization techniques, little attention has been paid to the scalability and effectiveness of these techniques for large programs. Although many optimizations have been shown to be effective for greatly improving the performance of moderately size programs, the increasing complexity and size, as well as the modular design of applications make the optimization techniques too costly or not effective.
The goal of this research project is to develop and experimentally verify a scalable approach to compiler optimizations. Instead of simultaneously considering optimization opportunities across the whole program, as current techniques do, program paths are used to limit the scope of optimizations. A demand driven framework for analyses and code transformations forms the foundation for the development of the techniques. Benchmark programs from government, industry, and standard benchmark suites are targeted to test the scalability of techniques developed.
Scalable techniques will be useful in industry to improve the performance of large programs, and the optimizer developed will serve as a platform to enable researchers to experimentally evaluate the scalability of optimizations. The demand driven analysis algorithms will not only be useful for scalable optimizations, but also for software engineering tools.
