| With multi-core processors becoming mainstream, multi-core programming standards are increasingly paid wide attention in recent years. There are two multi-core programming standards that are current mainstream: OpenMP and MPI. Compared to MPI which is multi-core programming standards oriented to message passing system architecture, it is much easier and more convenient to use OpenMP. The multi-core programming standards oriented to shared memory system architecture represented by OpenMP are developing rapidly along with the popularity of multi-core technology. How to effectively utilize OpenMP and farthest dig the potential parallelism in programs, as well as how to explore a series of optimization methods for OpenMP programs has a very strong practical significance and value for the development of future multi-core technology.The research work mainly focuses on two aspects below:(1) The research background of this paper is based on FT multi-core DSP. Starting from the parallel computing theory, we study the OpenMP programming model and characteristics, and put forward parallel optimization paradigm based on OpenMP for development flow of OpenMP programs, and summarize optimization methods and strategies in the process of developing OpenMP programs. Combined with UTDSP Benchmark Suite, these optimization methods are tested on FT multi-core DSP platform to verify the validity.(2) FFmpeg, an open source H.264 decoder, is implemented on FT multi-core DSP platform. By using analysis tools, hotspots and modules in this program are analyzed and optimized combined with OpenMP characteristics, and then optimization scheme is designed and implemented based on OpenMP. Finally, the optimized program has been tested on FT multi-core DSP platform, and the performance of H.264 decoder has been improved rapidly and effectively by achieving 20% boost nearly, which also verifies the OpenMP parallel optimization has the advantage of efficiency and convenience. |
PDF Full Text Request