OpenMP tasks: Asynchronous programming made easy

The task technology is one of most promising to exploit node parallelism for next generation of HPC architectures, like those under development within the CORAL initiative in the US. In general, single FPU performance is not going to increase any longer in contrast to single node performance. As a consequence, nodes architecture will feature more and more FPUs (proportional to the peak power of the node itself) and determine the need for an efficient technology able to take advantage of intra-node parallelism, reducing synchronization among threads and improving data locality to meet the memory hierarchy. In this paper we present a case study of node parallelization using SHYFEM code, a software for coastal area studies. To the best of our knowledge, this is one of the first work that shows how to fully parallelize a software with OpenMP task technology. We present the comparison between tasks and threads OpenMP programming paradigms, showing the advantages of using the hierarchical and asynchronous programming paradigm made available by OpenMP tasks.