TBFMM: A C++ generic and parallel fast multipole method library

Abstract

TBFMM, for task-based FMM, is a high-performance package that implements the parallel fast multipole method (FMM) in modern C++17. It implements parallel strategies for multicore architectures, i.e. to run on a single computing node. TBFMM was designed to be easily customized thanks to C++ templates and fine control of the C++ classes' inter-dependencies. Users can implement new FMM kernels, new types of interacting elements or even new parallelization strategies. As such, it can be used as a simulation toolbox for scientists in physics or applied mathematics. It enables users to perform simulations while delegating the data structure, the algorithm and the parallelization to the library. Besides, TBFMM can also provide an interesting use case for the HPC research community regarding parallelization, optimization and scheduling of applications handling irregular data structures.

ExaFMM: a high-performance fast multipole method library with C++ and Python interfaces

“…It comes with an option to use StarPU runtime system to handle heterogeneous task scheduling. TBFMM (Bramas, 2020) is a task-based FMM library that features a generic C++ design to support various types of tree structures and kernels, through heavy use of C++ templates. PVFMM (Malhotra & Biros, 2015) can compute both particle and volume potentials using a kernel-independent FMM, KIFMM (Ying et al, 2004).…”

Effective Implementation of the Fast Multipole Method for Particle Interaction with Newtonian Potential

“…Отдельно упомянем существующие открытые программные реализации быстрого метода мультиполей [14][15][16][17][18]. В данной работе мы проведем сравнение с двумя реализациями [14][15][16][17][18]. Дальнейшее изложение устроено следующим образом.…”