Energy Efficiency Research of Plasma Physics Codes

Modern trends in energy-efficient economy pose new challenges to the high-performance computing industry. Maximizing flops per watt value became one of the most important processes during code optimization. Considering this, we need to write our scientific codes, taking into account all modern hardware features. In this work, we present the energy efficiency research of our two plasma physics codes based on a particle-in-cell method for two independent physical problems, such as a diamagnetic trap for thermonuclear synthesis simulation and laser-plasma interaction problem. We used loops and data structure optimizations for Intel Fortran auto- vectorization. Most computationally complex loops did not use data that could be changed outside the loop, and mathematical operations were simplified for the automatic generation of FMA instructions by the compiler. We achieved a 30% energy efficiency boost for the diamagnetic trap and 45% for laser- plasma interaction PIC codes. This work was supported by the Russian Science Foundation (project 19-71-20026).