基于规范划分集的并行循环编译框架

1 引言并行循环描述大多数先进科学应用的核心部分,是颇有价值的并行性来源。高级并行语言往往提供专门指导语句表达并行循环,以便并行编译器利用并行循环的非数据相关特性。在数据并行程序设计语言HPF中规定,可以将关键词INDEPENDENT加在Do循环前面,指出其后的Do循环是一个并行循环。与之相关的编译实现也成为HPF编译器的一个重点。并行循环L可以严格定义如下:假设R(i)表示迭代i所

A Compiling Framework Based on Uniform Partitioning Schemes for Parallel Loops

Compilation of parallel loops is one of the most important parts in parallel compilation and optimization. This paper mainly discusses the key techniques during the compilation implementation of parallel loops, based on the uniform partition schemes. It includes techniques in local array index generating, loop space reconstructing, communication detecting and organizing and data dependence disposing. The efficiency of this implementation has been proved by lots of experiments. The p_HPF compiler which adopts this compiling framework can obtain good speedups and efficiencies. The compiler has been applied in many fields, particularly the field of petroleum exploration.