This project will develop link-time and run-time techniques for improving the performance of scientific programs. The overall goal is to increase the performance of both existing and new codes, whatever the source programming language, algorithm, application, or target architectures.
This project considers how tools for link-time and run-time optimization of advanced scientific programs can be constructed. Specifically, the following topics are investigated:
1. Link-time program analysis. The project will develop techniques for efficient and useful whole-program analysis, aimed at focusing the analysis effort on program components whose analysis and/or optimization can contribute to tangible improvements in program performance.
2. Whole-program optimization at link time. The project will investigate optimizations of executable programs, using classical techniques such as inlining applied to library code, as well as optimizations aimed at exploiting specific characteristics of the underlying hardware.
3. Run-time optimization. The project will also investigate optimizations that depend on values that are computed while an application executes; these include optimizations that effect the amount of concurrency in a parallel application as well as how communication and synchronization are realized.
