Parallelization For Serial Programs And Its Application On Desktop Grid

With the rapid development of computer software and hardware, high-performance computing has been gradually applied to an increasing number of fields. Parallel computing is an important technical way to achieve high performance. Parallelization for serial pro-grams, a significant method of parallel computing, can be mainly divided into three parts: dependency analysis, alias analysis, and data decomposition. Based on the researches on parallelization for serial programs, the main research work of this paper is as follows:Firstly, dependency analysis is the foremost task of parallelization. Dependency can be divided into two categories:control dependency and data dependency. In the step of control dependency analysis, this paper presents an improved algorithm, which analyzes inter-procedural control dependency by the conception of the control transfer caused by function calls and returns. In the phase of data dependency analysis, we give the algorithms for removing inter-statement and inter-iteration false dependencies.Later, we apply the algorithms to a parallel division model algorithm to verify the correctness of the algorithms.Secondly, alias analysis is an important phase, which impacts the degree of paralleliza-tion. In this paper, we present an algorithm for flow-insensitive alias analysis, which adds certain rules of updating alias set to improve the precision of alias analysis.Thirdly, the paper analyzes the issues of memory detection in semiconductor industry, and the usual solutions. On this basis, desktop grid is introduced into memory detection computing. Then, the idea of memory detection based on desktop grid is proposed, and an algorithm of data decomposition is given to improve the overall efficiency.The researches and improvements on the algorithms of parallelization have certain practical significance. Firstly, it is of referential value for dependency and alias analysis, which promotes further research of parallelization. Secondly, we give an alternative solu-tion to memory detection, which provides reference for other similar applications.