Gpu-based Hardware Acceleration In The Particle Simulation

With the rapid development of computer performance, people hope to simulate the characteristics of plasma more accurately, so this the whole three-dimensional relativistic particle simulation model is gradually raised, with the introduction of the macro particles up to 106 ~ 1010, which makes the simulative consumption of computer resources of the PIC (including the memory and computing time) increase in order of magnitude. In order to solve the huge computational burden, it's so vital to develop the corresponding parallel PIC simulative algorithm. Many mature methods of the parallel PIC simulative algorithms are based on the traditional central processing unit (CPU), but the limited resources of CPU become the bottleneck. Over the past decade, a new massive parallel processor - GPU (Graphic Processing Unit, graphics processor) began to be applied to scientific computing.The first chapter of the thesis describes the profile and the development of the simulation of plasma particular and the problems that faced in calculation. The developing course of the GPU and its progressing situation that applied in parallel computing.The second chapter introduces CUDA unified computing framework, in which the GPU is considered as a hardware and software systems of data parallel computing device. And in this part, the detailed description of the coarse-grained and fine-grained methods used when facing the computing task in the process of problem analysis is given, and also the mapping method and scheduler implementation in the GPU hardware of the large-scale parallel programme within the CUDA model.The third chapter introduced the simulative methods of the plasma particular, using the movement track of the simulative annular electron beam in the cylindrical system as an example. The theoretical analysis includes: the solution of electronic motion equation, the solution of the charge source and the solution of the three parts of the Poisson equation. As the electron beam has this ringlike feature and the axial symmetry of the cylindrical system, the cylindrical coordinates is applied in theoretical analysis and numerical simulation.The fourth chapter combines the introduction of the second and third chapters, analyzed the feasibility of the PIC simulation in GPU, brought forward the overall implementation process. The whole calculation includes: the solution of movement of charged particles , the updates of electron potential grid and the solution of charge density. In order to avoid the conflicts in memory while inputting, the algorithm may changed to atomic function when calculating the electron density. Providing the parallel version of data and the implementation form of CUDA, and also with the comparison of the serial programme of CPU, so a higher precision and a 10 times speed-up ratio is got.