Run-time Parallelization Optimization Techniques ? 1 Run-time Parallelization Requires Compiler Analysis

In this paper we rst present several compiler techniques to reduce the overhead of run-time parallelization. We show how to use static control ow information to reduce the number of memory references that need to be traced at run-time. Then we introduce several methods designed speciically for the parallelization of sparse applications. We detail some heuristics on how to speculate on the type and data structures used by the original code and thus reduce the memory requirements for tracing the sparse access patterns without performing any additional work. Optimization techniques for the sparse reduction parallelization and speculative loop distribution conclude the paper. Current parallelizing compilers cannot identify a signiicant fraction of paralleliz-able loops because they have complex or statically insuuciently deened access patterns. To ll this gap we advocate a novel framework for their identiica-tion: speculatively execute the loop as a doall, and apply a fully parallel data dependence test to determine if it had any cross{processor dependences; if the test fails, then the loop is re{executed serially. While this method is inherently scalable its practical success depends on the fraction of ideal speedup that can be obtained on modest to moderately large parallel machines. Maximizing the resulting parallelism can be obtained only through a minimization of the run-time overhead of the method, which in turn depends on its level of integration within a restructuring compiler. This technique (the LRPD test) and related issues have been presented in detail in 3, 4] and thus will not be presented here. We describe a compiler technique that reduces the number of memory references that have to be collected at run-time by using static control ow information. With this technique we can remove the shadowing of many references ? A full version of this paper is available as