Abstract: Minimizing communication by increasing the locality of data references is an important optimization for achieving high performance on large-scale parallel machines, but reorganization is inevitable in the progress of decomposition. It is important to find a method to solve the problem. This paper describes a linear decomposition algorithm which finds computation and data decomposition automatically, including finding data and computations decomposition that has data reorganization communication.

Key words: Paralleling compiler, Data decomposition, Computation decomposition, Data reorganization, Loop level parallel

摘要： 对于分布内存体系结构的并行计算机而言，如何对计算和数据进行合理划分以增加数据本地化减少处理器间的通信是提高其并行性能的关键，但在数据划分过程中，重分布通信有时不可避免，如何进行合理的数据和计算划分以减少通信并最大限度的利用程序的并行性是并行编译中的一个重要问题。该文主要讨论了一种支持数据重分布的自动进行计算和数据划分的算法。

关键词: 并行编译, 数据分解, 计算划分, 数据重分布, 循环级并行

CLC Number: 

• TP314