Scalable Hierarchical Multipole Methods Using an Asynchronous Many-Tasking Runtime System

Hierarchical Multipole Methods (HMMs) are an important class of methods in scientific and engineering applications. They are challenging to parallelize for contemporary and emerging platforms using existing programming models. Asynchronous many-tasking (AMT) execution models provide abstractions suitable for HMMs and promise scalability in the context of future exascale systems. In our work we (1) present a novel AMT-based HMM framework on HPX-5---an experimental exascale interface and runtime system, (2) evaluate its performance and scalability for a range of interaction types and data sets, and (3) reflect on the costs and benefits of using HPX-5 for this class of applications.