Development of efficient programs on GPU

Product data parallel GPU processor has recently attracted many application developers attention. GPU architecture now has many advantages. It can provide easier programmability and increase generality. GPU maintains the tremendous memory bandwidth and computational power which make it better than traditional CPU in doing computational problems. For this project aims to develop efficient programs running on graphical processor unit (GPU). The project involves GPU programming and testing. This project presents the test graphics processing unit (GPU) after make a comparison between CPU and GPU in the same environment. In this way we can determine the computational efficiency of GPU over CPU. The test method is about accelerated FADI-FDTD which is fundamental alternating-direction-implicit finite-difference time-domain with CFS-CPML (complex frequency shifted convolution perfectly matched layer. Using CUDA architecture to program GPU for CUDA is both a hardware and software platform that enables NVIDIA GPU to execute programs written with C/C++ or other languages. The FADI-FDTD with CFS-CPML is further incorporated into the GPU to exploit data parallelism. Results show that GPU can gain a much higher efficient.Bachelor of Engineerin