Asynchronous MPI Communication with OpenMP Tasks - Spawning Task Dependency Graphs Across Nodes

Explore asynchronous MPI communication with OpenMP tasks in this informative seminar from the NHR PerfLab series. Learn how to improve scalability of parallel codes by replacing block-synchronous execution with fine-grained synchronization using OpenMP tasks and dependencies. Discover the potential of detached tasks introduced in OpenMP 5.0 and their combination with MPI detached communication to build task dependency graphs across MPI processes. Gain insights into integrating MPI detached communication into your projects for real asynchronous communication benefits. Compare parallel performance of different synchronization levels through example code demonstrations. Understand how this approach can also be applied using C++ futures/promises for those not using OpenMP tasks. Presented by Joachim Jenke, a postdoctoral researcher from RWTH Aachen University, this seminar offers valuable knowledge for HPC application developers seeking to enhance correctness and performance.

Date and time: Tuesday, April 4, 2:00 p.m. – p.m. CET