Simulation of plasma dynamics and laser interaction with various targets is a high-demand area of computational physics. Among important applications are compact sources for hadron therapy for cancer treatment, bio-imaging, devices for research of intra-molecular and intra-atomic processes. Due to highly nonlinear effects and geometric complexity plasma simulation is often based on Particle-in-Cell (PIC) method. The distinguishing feature of this method is handling two distinct sets of data: ensemble of charged particles described by position, momentum, charge, and mass, and grid values of the electromagnetic field. Lack of straightforward ways of localizing memory access pattern causes challenges of efficient implementation for both traditional and heterogeneous cluster systems based on accelerators. Applications of the Particle-in-Cell method require supercomputers: some tasks require simulation of 109 particles and 108 grid cells.
Since 2010 our team develops electromagnetic PIC code PICADOR for heterogeneous cluster systems with CPUs, GPUs and Intel Xeon Phi. The code is among the first implementations optimized for Xeon Phi.
The project is developed in Lobachevsky State University of Nizhni Novgorod (UNN) in collaboration with researchers from the Institute of Applied Physics of the Russian Academy of Sciences (IAP RAS), Chalmers University of Technology and Umea University (Sweden).
Simulations are currently performed on the following supercomputers: Lobachevsky (UNN), Lomonosov (MSU), MVS-100K & MVS-10P (JSC RAS), Abisko (HPC2N, Sweden), Triolith (NSC, Sweden).
PICADOR is a tool for three-dimensional plasma simulation for traditional and heterogeneous supercomputers. It is based on the Particle-in-Cell method and has the following features: