CSR:Small:New Slicing Techniques for Program Parallelization

Purdue University, West Lafayette IN

Investigators

Abstract

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5). Multicore processors have become main-stream. At the same time, CPU clock rates are no longer increasing. Operating within this environment, application developers are under competitive pressure to parallelize software in order to achieve an aceptable performance level for computationally intensive features. Unfortunately, fully automated parallelization is constrained by sequential semantics and by limits of a compiler?s analytical abilities. Consequently, designing parallel programming languages for developing efficient yet reliable main-stream applications continues to be a challenge. Rewriting software completely by hand is undesireable, because this kind of activity may waste years of investment in large scale software. Therefore, ideally, existing software should be automatically converted to new parallel forms. To address this problem, new program slicing techniques are investigated in this project to enable programmers to convert existing sequential programs, with minimum hand-made changes, into forms which can be safely parallelized by automatic tools. This research will result in a set of novel program analysis and transformation techniques to support the new slicing methods. The project will have a broad impact on the US software industry?s ability to compete globally in its endeavor to retrofit existing software for the emerging hardware platforms. The tool and techniques developed in the project will be used in compiler and architecture courses offered at both graduate and undergraduate levels. The project also engages both graduate and undergraduate students in the key research activities, providing advanced technical training which will be critical to future success of a new generation of computer scientists and software engineers.