A Method of Computation Decomposition on Tightly-Nested Loop Automatic Parallelization

An automatic parallelization method for tightly-nested loops running on multi-core system has been proposed. First, according to the physical characteristics of multi-core processors, a way has been presented to solve the problem on dada locality during data decomposition; Second, for increasing parallel granularity of tight nested loops, the method discussed in this article studied computation decomposition based on workload, and brought forward how to compute the workload of loop iteration that can be run in parallel, and at last according to the size of the workload, determined the granularity of parallel loops to achieve to reduce the parallel overhead brought by the parallel iteration of small workload. Using this method, an automatic parallelization model based on workload can be constructed.