Organization And Application Research Of Data Structure In GPU General Purpose Computation

The research of computer graphics processor unit (GPU) used for general purpose computation has become a hot field. GPU data structure organization and application is the premise to study GPU computing. Without the assistance of data structure, no realization of the specific operation would be achieved. This thesis is based on the analysis of the GPU architecture, programming interface, the common shade language and the flow programming model, choose linear table the basic data structure as the study object. Detailed analysis was made on physical and logical structures, the organize method was presented based on it. A solution of the linear table delete and insert operations was proposed under the circumstances in GPU through comparative analysis and comparison.The dissertation applies this data structure into enumerate search and discusses the general-purpose computing of GPU. The improvement in software was introduced due to poor CPU performance of the same search, a parallel realization was presented. Studied the hardware improve process, which is the process of how to transplant exhaustive search to GPU environment. The performance testing of enumeration method of the CPU and GPU was compared. The specific performance metrics were achieved from the analysis of the test results.Finally, we reached the following conclusions: without the consideration of the bandwidth influence, the more of the data scale the better speedup will be achieved of GPU; The problems in which the shade algorithm can be called many times are very suitable for operating under the GPU. This can increase the proportion of the kernel algorithm in time spending, thereby enhancing operational efficiency.