Advanced Heterogeneous Programming on GPUs with MPI & OmpSs

The students who finish this course will be able to develop benchmarks and simple applications with the MPI/OmpSs programming model to be executed in clusters and clusters of GPUs. They will be also able to implement optimizations in their own codes.

Session 1: Introduction to OmpSs. Intra-node parallelism, task-model. It includes specific hands-on exercies.
Session 2: Introduction to MPI/OmpSs . Inter-node parallelism, interoperability. It includes specific hands-on exercies.
Session 3: Introduction to OmpSs at GPUs. Device offload and libraries. It includes specific hands-on exercies.

• Xavier Martorell
• Xavier Teruel

This tutorial addresses the growing need for portable and efficient parallel programming models capable of delivering high performance on modern HPC systems, including large-scale clusters and GPU-accelerated platforms. The course introduces the hybrid MPI/OmpSs programming model as a practical and forward-looking solution for future exascale systems. The session begins with an overview of the model’s conceptual foundations, including task-based parallelism, dependency management, devices offloading and its integration with message-passing paradigms.

Good knowledge of C/C++
Basic knowledge of CUDA/OpenCL
Basic knowledge of Paraver/Extrae

Certificate of Achievement

laptop + ssh connection