Abstract:
A high performance implementation of particle-in-cell methods for laser plasma simulation is considered. The PICADOR code is used. An efficient utilization of the new Intel Xeon Phi coprocessors is discussed. It is shown that a code optimized well for traditional CPUs is not always efficient on coprocessors without additional optimization. A number of ways for the performance optimization of numerical plasma simulation are analyzed. The results of computing experiments show 1.8 times speed up on Xeon Phi compared to an optimized code on CPU.