Testing the Effectiveness and Efficiency of Data Dependence Decision Algorithms

点击翻译按钮获取中文摘要

Data dependence is the theoretical cornerstone on which all automatic parallelism detection methods are built. Automatic vectorization, which has been critical in the use of vector computers, and other advanced optimizations for new parallel computers and languages all require data dependence information. Our project will measure both the effectiveness and the efficiency of data dependence decision algorithms. Effectiveness will be measured by counting the number of loops that can be vectorized and concurrently executed. Efficiency will be measured by computing the time taken to build the data dependence graph. Efficiency must be carefully defined, since a quick data dependence decision algorithm may add more arcs to the dependence graph, thus slowing down other phases of the compilation which depend on the number of arcs in the graph.