GPU Acceleration Of 1D3V Particle-In-Cell Simulation Software BUMBLEBEE

Energy plays a critical important role in the daily production and lives of human beings. With the rapid development of the society,the human society has an increasing demand for energy, so the energy has become the problem needed to solve by all the countries now.The process of nuclear fusion can produce massive energy,and the environmental pollution caused by nuclear fusion is small. At the same time, the raw materials used in nuclear fusion are Deuterium and Tritium, which are so rich in the sea water, so nuclear fusion is a key research subject in the world. In the inertial confinement fusion,the fast ignition has a low requirements of driven compression symmetry, and it could use less energy to achieve ignition and high gain of the fusion,so the fast ignition become one of the hotspots in the research of inertial confinement fusion. Due to the complexity and difficulty of the experiment, the computer simulation has become one of the important means to study the fast ignition. Particle simulation method is one of the main methods, but it will generate a huge computational burden. With the development of the CPU tending to be saturated, the GPU hardware acceleration of the particle simulation software will make it possible to study the particle simulation of the key physical problem efficiently.This paper is based on the Particle-In-Cell simulation software BUMBLEBEE developed by UESTC, we paralleled and optimized the part of interaction between laser and plasma and developed the paralleled 1D3 V Particle-In-Cell simulation software BUMBLEBEE based on GPU. The main work done as follows:1.Summarized the basis of the CUDA parallel computing, and mainly introduced the CUDA memory, CUDA software architecture and the implementation of the CUDA execution mode.2.We did a overview for the serial Particle-In-Cell simulation software BUMBLEBEE from the function module, simulation process and the data structure used in the software. And at last, we initially paralleled the BUMBLEBEE software of the part of its particle motion, the source solution and the field solution based on GPU.3.We proposed the GPU parallel program performance optimization strategy, and then optimized the paralleled BUMBLEBEE from the thread scheduling, shared memory and the registers combined with the actual situation of the parallel program. Finally, we presented the results of the parallel programs after the performance optimization.4.Compared the simulation results of the serial and parallel BUMBLEBEE software to verify the correctness of parallealization. And we analyzed the simulation results of the particle density distribution, field changes in detail, and finally tested a large number of different interaction examples to give a detailed speedup of the test results.