Mobile GPU Parallel Particle Swarm Optimization Algorithm Based On Compute Shader

Particle Swarm Optimization(PSO)algorithm belongs to cluster intelligent algorithm.PSO has such a strong global search and convergence ability together with a simple and easy concept to implement that it develops rapidly into a large range of areas.However,most problems in reality are discretized and diversified,which leads to the relatively low speed of the algorithm.Fortunately,particles in PSO do not interfere with each other that the parallel acceleration program can serve as a good solution.The rapid development of GPU technology brings the upsurge of research and application of GPU-based general-purpose computing(GPGPU)in high performance computing.At present,main GPGPU tools include NVIDIA's CUDA,Apple's OpenCL and so on.Compute Shader is new features of Microsoft in DirectX 11 API,which can also be done on the GPU parallel computing.Although Compute Shader in Unity ids built on the DirectX 11 DirectCompute technology,since the Compute Shader is introduced in the OpenGL ES 3.1 that it is possible to use the Compute Shader as a GPGPU on the mobile side.In order to solve that the PSO algorithm is relatively low when facing complex problems,the parallel particle swarm optimization algorithm based on Compute Shader on the mobile GPU is under research.The main research results include:(1)Comparative analysis of three existing CUDA-based parallel PSO algorithm about its advantages and disadvantages.(2)Based on the analysis of the traditional PSO algorithm and the GPU-based parallel computing technology which is widely used at present,a parallel particle swarm optimization algorithm based on Compute Shader is designed and implemented.By using a large number of threads in mobile GPU to parallel process the search work in different dimension of each particle,it can speed up the convergence rate of the entire particle group.Analysis of the test results for multiple reference functions shows that compared with the CPU-side serial PSO algorithm,method in the present study can achieve maximum of 250 times in speed ratio.