The International Conference for High Performance Computing, Networking, Storage and Analysis
Task Profiling through OpenMP Runtime API and Tool Support.
Authors: Ahmad Qawasmeh (University of Houston), Abid M. Malik (University of Houston), Deepak Eachempati (University of Houston), Barbara M. Chapman (University of Houston)
Abstract: The introduction of tasks in the OpenMP programming model brings a new level of parallelism. This also creates new challenges with respect to its applicability through an event-based performance profiling. The OpenMP Architecture Review Board (ARB) approved an interface specification known as the "OpenMP Runtime API (ORA) for Profiling" to enable performance tools to interact with OpenMP programs.
With negligible overheads, we propose new extensions to the ORA for profiling task-level parallelism. We implement these extensions in the open-source OpenUH compiler. We present an efficient method to distinguish individual task instances to capture their associated events. These events illustrate the behavior of OpenMP tasks and identify the overheads associated with the OpenMP tasking model.
Additionally, we integrate our ORA into the performance tool TAU to visualize task measurements at the construct-level. The proposed extensions are in line with the specification recently proposed by the OpenMP tools committee for task profiling.