Research And Implementation Of Parallel Atmi Simulator On GPU Platform

At present, the growing of chip temperature has become one of the main bottlenecks in microprocessor chip design. The most commonly used methods for solving the heat equation are Finite Difference (FDM) and Finite Element Methods (FEM). These two methods permit modeling arbitrarily detailed physical systems. However, using FDM or FEM to model microprocessor temperature requires a large number of nodes and a long computation time.The ATMI simulator solves the heat equations using an analysis method (AM) The advantages of analytical method are the reduced computing nodes and a fast computing speed. ATMI simulator has relative fast computing speed, high reliability, and easy to use. These advantages make the ATMI simulator widely used in academic research. Experiments show that the thermal response stage takes a long time when to simulate more than a dozen cores of microprocessor. With the development of the future processor, ATMI is no longer suitable for the simulation of future processor temperature.In this thesis, we implement the ATMI simulator on GPU platform with CUDA. It mainly includes the following aspects of work:First, we investigate the serial algorithm of ATMI simulator in-depth and focus on the algorithm in thermal response phase, then combining with GPU architecture and CUDA programming characteristics, we optimize the serial algorithm in the original thermal response phase. Secondly, we realize numerical integrate algorithm and Bessel algorithm on the GPU platform. Since thread parallelism is not enough and complex structure of the program, the experimental results of two algorithms are not satisfactory compared with the original serial algorithm. Finally, we present and design parallel algorithm of heat response phase on the GPU the platform and completed the parallel algorithm frame. At present, the result of the parallel programs and the original serial program results is not consistent, and some problems and difficulties need to be further solved.In this thesis, we focus on the research and exploration of simulator on the GPU platform, and want to solve some issues of the GPU multi-threaded applications and accumulate experience in the related problems.