Explore the vast range of titles available.

This chapter describes how OmpSs extends the OpenMP 3.0 node programming model and how it leverages message passing interface (MPI) and OpenCL/CUDA, mastering the efficient programming of the clustered heterogeneous multi‐/many‐core systems that will be available in current and future computing systems. It describes the language extensions and the implementation of OmpSs, focusing on the intelligence that needs to be embedded in the runtime system to effectively lower the programmability wall and the opportunities to implement new mechanisms and policies. The chapter reasons about the overheads related with task management (detecting intertask data dependencies, identifying task‐level parallelism and executing tasks out of order) in OmpSs examining how far a software implementation can go to cope with fine‐grain parallelism and opening the door to novel hardware mechanisms for emerging multicore architectures. The chapter provides a brief description of the OmpSs execution model to understand the programming model extensions.