The Improvement And Research Of DEM Based On OpenCL

The traditional GPU programming support for non-graphics applications limitedonly by the graphical interface programming, which requires the programmer has aprofessional background graphics. In addition, traditional GPU does not providesupport for double-precision floating point. Therefore, GPU in scientific computingapplications often limited.In recent years, GPU-based general-purpose computing (GPGPU) has achieved amajor breakthrough, GPU manufacturers constantly GPU programming modelprovides improved so easy programming of the GPU continues to improve, Such asNVIDIA has introduced CUDA (Compute Unified Device Architecture) programmingmodel, The Brook+language of AMD company and Apple’s OpenCL language.There are two main GPU programming languages: Graphics API, such asDirectX and OpenGL, GPGPU languages such as OpenCL and CUDA. GPGPU is theGPU as multicore processor architecture, offers a variety of language class interface,will be exposed to the general-purpose computing hardware capabilities.OpenCL was originally developed by Apple in recent years, with AMD, IBM,Intel and NVIDIA OpenCL technical teams to collaborate on a preliminary perfect.OpenCL is a programming framework for heterogeneous platforms, OpenCL is anopen source programming language, the algorithm can be written in various types ofGPU OpenCL, and OpenCL has good compatibility, is conducive to the transplantprocedure.In reality, there are a lot of particulate matter, such as dust, rocks, plants andseeds, filled with all kinds of particulate matter in the field of production of humanlife, so the study of particle systems help people understand the world, to improve andimprovement of human life and production. Because now the bottleneck in thedevelopment of computer hardware, making large-scale computing power of the CPUfor data calculation somewhat stretched. However, the emergence of GPU parallelcomputing allows people to study particle system has entered a new era. GPU has not weaker than the CPU’s floating point computing power, and far more than the GPU’sprocessor CPU, making the GPU’s parallel computing power more than CPU.Discrete Element (DEM) of the study was mainly targeted at the mechanicalbehavior of non-continuous media rock, etc. The basic idea is to discontinuitiesseparated into a collection of rigid elements, so that each rigid element satisfies theequation of motion, the method used when歩iteration to solve each equations ofmotion of rigid elements, then seek not the whole continuum of movement patterns,DEM permit relative movement between cells move, do not have to meet thecontinuous displacement and deformation compatibility conditions, calculation speed,the required storage space is small, especially suitable for solving large displacementand nonlinear problems.After years of unremitting efforts, our group theoretical foundation andarchitecture for discrete element of in-depth research, which also includes the study ofparticle systems. However, when the large number of particles operations, longrunning time of the program, resulting effect is far from ideal.In this paper, the algorithm based on the existing discrete element method based onparticle systems, through OpenCL algorithms, particle systems to improve the overallcomputing tasks into two parts: inter-particle collisions and particle collision with theboundary. Inter-particle collisions will be written using the OpenCL algorithm, mainlydivided between the particles in contact with the particles collide to determinemechanical calculations; the particles collide with the border is still done in the CPU,implementation and OpenCL parallel computing algorithms, ie CPU and GPU parallelcomputing, the main points to determine the particle boundary contact and collisionmechanics calculations. OpenCL optimization based particle system, making theoperational efficiency and effectiveness of the particle system can be improved.Through extensive testing program to see the optimized simulation results with theoriginal program is similar, but the running time is greatly reduced, proved OpenCLalgorithm to optimize the effect of the particle system played.In this paper, OpenCL memory fetch been optimized. Data exchange with thememory fetch OpenCL program is a major factor affecting the efficiency ofreasonable memory fetch mode can greatly enhance the OpenCL program hadoperating efficiency. In this paper, the kernel function data were systematicallyplanned, depending on the type of data, will use a different memory and for thememory fetch some of the issues raised reasonable solution to ensure fetch reduce overhead, on the other hand OpenCL program to enhance operational efficiency.Finally, the application has been fully tested by borderless test accuracy andoperational efficiency of inter-particle collisions OpenCL conducted a comprehensiveanalysis algorithm; border through testing, inter-particle collision and running on theGPU on the CPU particles colliding with the boundary running accuracy andoperational efficiency of parallel computing a comprehensive analysis.