Extending The Openmp Standard For Thread Mapping And Grouping

In this paper, we are exploring the idea of improving the OpenMP 2.5 standard to facilitate parallel programming on emerging architectures. This includes mapping threads to particular processors, improving the load balance among processors by work distribution, and supporting nested parallelism inherited from applications. We will demonstrate the performance gains of thread mapping with our experimental implementation, and propose new concepts in the standard to try addressing the issue in a broader sense.