Parallel optimization research based on numerical simulation

In this paper, we design a frame to automatically select CPU or GPU environment and establish four general parallel foundation libraries. First, the system general foundation library and mathematical foundation library are built on basis of the research on GPU parallel characteristics, CUDA programming technology and the serial algorithm of forward and preserved amplitude evaluation. Then, according to these two support libraries, we establish the forward general foundation library and preserved amplitude evaluation general foundation library, following several optimization strategies and criterions. Finally, accuracy and time experiments are taken to evaluate the two types of general parallel foundation libraries. It is showed the speedup has reached more than 10 times under the same accuracy.