Research Of X10 Language Mechanisms And Optimization

As multi-core architecture has become the mainstream of the development of the processor technology, parallel programming attracts the public attention. It is indispensable to investigate new parallel programming methods in order to utilize the advantage of multi-core processors. Traditional parallel programming languages are only used in certain domains; many of them describe the beneath details of parallel programming and are difficult to used for parallel programming, which leads to the poor efficiency of parallel applications'development. In order to utilize and popularize parallel programming, many new parallel programming models and languages turn up.In this paper, we investigate the X10 programming language. It is developed in the HPCS project for high performance computing; and it is an object-oriented language and aims enhance the efficiency of parallel programming and provide supports for high performance,portability and robustness. Our main contributions are as follows:1. Investigate the language mechanism of X10. By contrasting with current mainstream programming languages(such as OpenMP, MPI),other HPCS languages and new programming language---Go, find the problems in the X10 language.2. Programming with X10. Implement the tracer interface moving grid method YGX program with X10 and test the correctness of the implemented YGX program. Compare the implementation and the performance of X10 with the YGX program implemented by OpenMP.3. Optimize X10 language. Design and implement an algorithm based on program characteristics which is used for unrolling times; and implement the loop unrolling method for loop optimization. The result of the experiment shows it really improves the performance of the YGX program implemented with X10 about 10 percent.4. Investigate the technique for optimizing the X10 Parallel recursion programs. By restricting the number of spawning, reduce the influence on the performance of recursion programs. The experiment shows that restricting the numbers of activity properly in the parallel recursion programs can enhance the performance of parallel recursion programs.