UPC共享访问消息向量化

THE MESSAGE VECTORIZATION OF THE UPC SHARED ACCESS

UPC(Unified Parallel C)原型系统是基于开放源码的Berkeley UPC1.0进行开发的,具有易编程、可移植的优点。但由于缺乏有效的并行编译优化支持,其性能与商业UPC编译器相比,有着较大的差距。为了提高UPC并行应用性能,研究并实现了UPC共享访问消息向量化技术。在8个双核CPU的硬件环境上的测试表明该优化效果非常好。

UPC （ Unified Parallel C） prototype compiler is developed by Berkeley UPC 1.0 based on open source with the advantage of easy programming and portability. However, there is a big performance gap between it and the commercial UPC compilers, for lack of effective parallel compiler optimization support. In order to raise the performance of UPC parallel applications, in this paper it studies and implements the shared access massage vectorization technique. The test results under the hardware environment of 8 CPUs （two kernels） show that the optimization has very good effect.