Reservoir Labs Presents “Analysis of Explicit vs. Implicit Tasking in OpenMP Using Kripke” at Supercomputing 2018
As part of the 2018 Supercomputing Conference in Dallas, TX, Reservoir Labs is presenting a Workshop entitled “Analysis of Explicit vs. Implicit Tasking in OpenMP Using Kripke“. Reservoir engineers Charles Jin and Muthu Baskaran will discuss Reservoir’s latest research on the advantages of explicit task-based parallelism for real-world applications where the pattern of dependence relations makes implicit parallelism inefficient. This is demonstrated through the sweep kernel of the Kripke benchmark. The kernel has high levels of parallelism, but it is only effective to access this parallelism in the context of a run-time that accepts and schedules with explicit dependences between tasks. The demonstration is in the context of OpenMP 4.5. The work motivates our R-Stream capabilities for generating such explicit task based parallel programs automatically from simple high-level specifications of the kernel.