Research And Implementation Of Source Code Parallelization Based On Multi-Cores

With the rapid and wide applying of multi-core CPU in two years, how to take full advantage of the ability of parallel processing for multi-core computers and program the parallel programs is the issue in the research and application field of multi-core.In this paper, we analyzed multi-core computing technologies and proposed a combined method to extract in-process and inter-process parallelizable information from the source codes. We used the international parallel benchmarks (such as Livermore, LAPACK, NAS, etc) as the experimental codes, and compared the results with the Intel Parallel Studio tools (IPS) to verify our method.Firstly, we provided a method of extracting the in-process parallelizable information which is based on the data dependencies of source codes. It applied SVM to mine the parallelizable points which used executing time, data dependency, depth of recursive and number of recursive as features. The experimental results show that the parallelizable points mined by our method are able to meet the needs of practical application, and it also can get considerable speed-up after parallelizing these points.Secondly, we proposed a method to extract the inter-process parallelizable points by constructing inter-process dependency graph (IPDG) and it's effective to combine the inter-process dependency information. We also found that the in-process dependency information could effect the inter-process dependency information, so we applied the in-process dependency information to optimize the inter-process dependency information. The experimental results show that our method can extract inter-process parallelizable information more effectively.Finally, we combined above methods and did experiments on the parallel benchmarks. The experimental results show that our combined method also can extract comprehensive and effective parallelizable points on parallel benchmarks.