分簇结构高效谓词机制及编译优化方法

Effective predicate mechanism for clustering and its compiling optimization method

谓词执行是在控制流存在的条件下可以有效挖掘指令级并行性的硬件机制,而在分簇结构上实现谓词机制,可以提高分簇结构上条件的执行效率。针对分簇结构展开谓词体系结构的研究,提出了分簇结构部分谓词的高效实现方法,以及基于循环展开的分簇结构部分谓词支持框架。实验表明,提出的分簇结构部分谓词及编译框架可以很好地提高条件执行程序的执行效率,部分典型算法核心的加速比可以达到350%500%。

Predicate execution is a valid method to develop instruction parallelism in the programs with control flow.And implementation of predicate for clustering can improve efficiency of conditional statement.Researching on predicate mechanism for clustering,this paper presented an efficient partial predicate mechanism for clustering,and proposed a compiling implementation framework based on loop unrolling for partial predicate.Experiments show that the predicate implementation framework can better enhance executive efficiency of programs with conditional statement.The speed for some classic algorithms kernels can reach 350%~500%.