Abstract:

The authors proposed an automatic parallelization method on tight nested loops  that is running on multicore system. The method can be used to solve two problems according to the physical characteristics of multicore processors,  one is  data locality existing in data decomposition and the other is computation decomposition based on workload, and brought forward how to compute the workload of loop iteration that can be run in parallel. According to the size of  workload, we can determine the granularity of parallel loops so as  to reduce the parallel overhead brought by the parallel iteration of small workload. An automaticparallelization model based on workload can be constructed based on this method.

Key words: computation decomposition, workload, data decomposition, data locality, loop parallelization