An Improving Computation and Data Decomposition Method
Data locality is critical to achieving high performance on high-performance parallel machine. So how to find a good data decomposition is becoming a key issue in parallelizing compilers. We have developed a compiler system that fully automatically parallelizes sequential programs and optimizes data decomposition to improve data locality. Our data decomposition algorithm consists of two steps. The first step chooses the basic data and computation decomposition without considering read-only data. The second step then changes the data decomposition considering read-only data.We ran our compiler on a set of application programs. The results show that the algorithm can effectively discovers parallelism.
data decomposition computation decomposition read-only data loop level parallelism
Chunli Dong Rongcai Zhao Jianmin Pang Lin Han Peng Du
National Digital Switching System Engineering & Technology R &D Center Zhengzhou 450002, P.R CHINA
国际会议
杭州
英文
263-266
2006-10-12（万方平台首次上网日期，不代表论文的发表时间）