程序并行化中基于暴露集生成数据分布代码  

Generating data distributing code based on exposed-set for program parallelization

作　　者：丁锐[1] 赵荣彩[1] 韩林[1] 

机构地区：[1]解放军信息工程大学,河南郑州450002

出　　处：《计算机工程与设计》2009年第15期3573-3576,共4页Computer Engineering and Design

基　　金：国家863高技术研究发展计划基金项目(2006AA01Z408)

摘　　要：在并行化编译中,代码生成属于编译器的后端,决定着并行程序的执行效率。数据划分将计算循环中被重定义或没被读引用的数据映射到处理器,按照数据划分生成通信代码会产生冗余通信。提出了利用数组数据流分析求解暴露集,并建立计算划分、循环迭代以及暴露集的不等式限制系统,最后通过FME(fourier Motzkin elimination)消元生成数据分布代码的优化算法。测试结果表明该算法对数据分布的优化效果明显。The code generation belongs to the backend ofparallelizing compiler. Efficiency of the codes determines the parallel programs＇. The data being redefined and read in computing loop also is mapped to processor by data decomposition. Generating data distribution code by data decomposition brings redundant communications. An optimization algorithm for data distribution code generation is pre- sented. The algorithm got the exposed-set based on the array data-flow analysis, then created an inequality system with computation decomposition, loop iteration and exposed-set, at last realized auto-generation of the data distribution code by using FME elimination method. The testing results prove that the algorithm improve on the efficiency of data distribution.

关 键 词：并行编译 代码生成 暴露集 不等式系统 通信优化 

分 类 号：TP314[自动化与计算机技术—计算机软件与理论]