Automatic and Interactive Parallelization Automatic and Interactive Parallelization

The goal of this dissertation is to give programmers the ability to achieve high performance by focusing on developing parallel algorithms, rather than on architecture-speciic details. The advantages of this approach also include program portability and legibility. To achieve high performance, we provide automatic compilation techniques that tailor parallel algorithms to shared-memory multiprocessors with local caches and a common bus. In particular, the compiler maps complete applications onto the speciics of a machine, exploiting both parallelism and memory. To optimize complete applications, we develop novel, general algorithms to transform loops that contain arbitrary conditional control ow. In addition, we provide new interprocedural transformations which enable optimization across procedure boundaries. These techniques provide the basis for a robust automatic parallelizing algorithm that is applicable to complete programs. The algorithm for automatic parallel code generation takes into consideration the interaction of parallelism and data locality, as well as the overhead of parallelism. The algorithm is based on a simple cost model that accurately predicts cache line reuse from multiple accesses to the same memory location and from consecutive accesses. The optimizer uses this model to improve data locality. It also uses the model to discover and introduce eeective parallelism that complements the beneets of data locality. The optimizer further improves the eeectiveness of parallelism by seeking to increase its granularity. Parallelism is introduced only when granularity is suucient to overcome its associated costs. The algorithm for parallel code generation is shown to be eecient and several of its component algorithms are proven optimal. The eecacy of the optimizer is illustrated with experimental results. In most cases, it is very eeective and either achieves or improves the performance of hand-crafted parallel programs. When performance is not satisfactory, we provide an interactive parallel programming tool which combines compiler analysis and algorithms with human expertise. Acknowledgments Ken Kennedy provided me with the three most important elements of support in graduate school: intellectual, political and nancial. In addition, Ken, Keith Cooper and Linda Torczon fostered a research atmosphere and working environment whose beneets are untold. I would also like to recognize the other members of my committee, has been an endless source of encouragement and wisdom throughout my graduate career. Don Johnson gave me my rst taste of research and hooked me for life. I am fortunate that many of my fellow graduate students and friends supported my research intellectually, emotionally and with implementations. I would …