Efficient parallelization for 3d-3v sparse grid Particle-In-Cell: Shared memory architectures
Fabrice Deluzet, G. Fubiani, Laurent Garrigues, Clément Guillet, Jacek Narski
Abstract
Particle-In-Cell (PIC) schemes are ones of the most broadly used numerical methods in kinetic simulation of plasmas. The contribution of the present paper is dedicated to the introduction of novel parallelization strategies specific to shared memory architectures tailored for Particle-In-Cell methods implementing sparse grid reconstructions. These strategies operate the different parallelisms genuine to Sparse-PIC methods to obtain speed-up exceeding 100 on 128 cores using a strong scaling study of a 3d-3v simulation. On top of that, substantial gains (two orders of magnitude) are introduced for sequential as well as parallel computation of the electric field thanks to the hierarchization procedure. It consists in decomposing the information carried by sparse grids onto hierarchical basis functions, entailing a significantly reduced number of operations.