Abstract:
Given the limitations of cost and flexibility in OS-level threads, the trend of utilizing the user-level threading (ULT) and tasking models becomes popular. However, the interoperability with existing high level programming models is always an issue that constrains the values and usability of user-level threading libraries. In the previous work, a lightweight, low-level tasking and threading framework named Argobots, is designed to efficiently overlap the communications and computations in parallel applications. In order to strengthen the interoperability of Argobots, BOLT is designed as a runtime library based on the Argobots to interoperate with OpenMP. In this work, the performance impacts of BOLT is evaluated and the potential problems in BOLT are further investigated and resolved to boost its performance and interoperability.

Short Bio:
Xi Wang is a Ph.D. student in the Computer Science Department at Texas Tech University (TTU), US.  He received his M.S. degree in Computer Science at TTU. His research area is High Performance Computing (HPC) with a focus on memory systems and computer architecture design. In particular, his current work is about boosting the performance of memory coalescing through advanced dynamic memory coalescing (DMC) techniques in HPC.