Section: New Results

Efficient Data Structures for a PIC Code on SIMD Architectures

Participants : Yann Barsamian, Éric Violard.

In collaboration with Sever Adrian Hirstoaga (mathematician researcher, member of Inria team TONUS), we have developed an efficient particle simulation code. The domain of application is plasma physics, the Particle-In-Cell code simulating 2d2v Vlasov-Poisson equation on Cartesian grid with periodic boundary conditions for Landau damping test-case. We first analyzed different strategies for improving its performance on single core and then we used a standard approach for parallelizing it on many cores using hybrid OpenMP/MPI implementation. The optimization of the sequential code is mainly based on (i) a structure of arrays for the particles, (ii) an efficient data structure for the electric field and the charge density, and (iii) an appropriate code for automatic vectorization of the charge accumulation and of the positions’ update. The parallelization of the loops over the particles is performed in a simple way (without domain decomposition) by means of both distributed and share memory paradigms. Satisfactory strong and weak scaling up to 8,192 cores on GENCI’s supercomputer Curie are obtained, bounded as expected by the overhead of MPI communications. A conference paper presenting this work is currently under review.