Heterogeneous Programming and Optimization of Gyrokinetic Toroidal Code and Large-scale Performance Test on TH-1A

Abstract. In this work, we discuss the porting to the GPU platform of the latest production version of the Gyrokinetic Torodial Code (GTC), which is a petascale fusion simulation code using particle-in-cell method. New GPU parallel algorithms have been designed for the particle push and shift operations. The GPU version of the GTC code was bench-marked on up to 3072 nodes of the Tianhe-1A supercomputer, which shows about 2x–3x overall speedup comparing NVIDIA M2050 GPUs to Intel Xeon X5670 CPUs. Strong and weak scaling studies have been per-formed using actual production simulation parameters, providing insights into GTC’s scalability and bottlenecks on large GPU supercomputers