| Particle system is one kind of system which uses a large number of particles to do simulation. This kind of system is wildly applied in the simulation of solid particles and fluid, including interactive process of fluid, smoke effects, simulation in astrophysics and molecular dynamics, and so on.The Discrete Element Method (DEM) is an important numerical method for particle system. The DEM particle system is applied in debris flow the solids-inflow zone of a single screw extruder, and so on. The DEM considers each particle in a granular assembly as a separate entity, which can interact with other particles or surfaces through collisions or lasting contacts. Use this method we can model a-dynamic behavior of a granular flow.This paper used CUDA to program a particle system based on DEM. The whole simulation system is divided into two parts. The host part was programmed in C++language. A particle system class was created as the framework of the simulation system. The simulation scene was draw by OpenGL. The device part was programmed in CUDA. The major undertaking is neighbor particles search, force computing of the particle, the physical quantities updating at the next time step.This paper tried many ways to improve the particle system's performance. Used a new neighbor particle search method which is more suitable for parallel program; Used shared memory to reduce the memory accessing in global memory; Used CUDA stream and asynchronous data copy to the transmission speed between the host and the device; Used OpenGL interoperability with CUDA to simplify the process of data transmission; The particle system also used program branch controlling, register strategy, the thread dimension division strategy and memory coalescing to optimize system performance. We used different graphic cards to run the particle system, and compared two GPU's performance. By comparing some particle system, we conclude the GPU-based particle system has better performance. |
PDF Full Text Request