Mastering Tasking with OpenMP 6.0

OpenMP is the leading portable, widely supported and easy-to-use directive-based programming model. OpenMP includes tasking support that enables composability and parallelization of a wide range of algorithms, including irregular ones. Tasking is now pervasive in OpenMP as it serves of the basis of all OpenMP parallelization constructs, including those for offloading code to attached accelerators. Mastering the tasking concept requires a change in how developers reason about the structure of their code and how to expose its parallelism. Our tutorial addresses this critical aspect by examining the tasking concept in detail and presenting solution patterns for many common problems.

With the recent release of OpenMP 6.0, tasking has been greatly extended. Most prominently, it now supports free-agent threads, which are threads that are not assigned to any team that is executing a parallel region, as well as task graphs for efficient replay. In this tutorial, we specifically present the new release.

We assume attendees understand basic parallelization concepts and know the fundamentals of OpenMP. We present the OpenMP tasking language features in detail and focus on performance aspects, such as introducing cut-off mechanisms, exploiting task dependencies, and preserving locality. All aspects are accompanied by extensive case studies.