类型转换语句的SLP发掘方法  

SLP Exploitation Method for Type Conversion Statements

作　　者：赵博[1,2] 赵荣彩[1,2] 李雁冰[1,2] 高伟[1,2] 

机构地区：[1]信息工程大学,郑州450001 [2]数学工程与先进计算国家重点实验室,郑州450001

出　　处：《计算机科学》2014年第11期16-21,共6页Computer Science

基　　金：“核高基”国家科技重大专项(2009ZX01036-001-001-2)资助

摘　　要：多媒体技术的迅速发展使得越来越多的处理器集成了SIMD扩展,当前的编译器大多数都已实现了自动向量化功能。为了发掘迭代内并行,一些编译器在自动向量化模块中引入了SLP向量化方法。多媒体数据的密集存储和规则运算使得在处理多媒体数据时需要进行频繁的数据类型转换,而目前的SLP向量化方法对数据类型转换的处理能力还不完善。为了在存在大量数据类型转换语句的程序中发掘更多的SLP向量化机会,提出了一种类型转换语句的SLP发掘方法,它能够在SLP向量化框架下利用数据重组实现具有相同向量化因子和不同向量化因子的数据类型之间的转换。实验结果表明,该方法能够有效地对类型转换语句进行SLP向量化发掘,提高了程序的向量化执行效率。With the rapid development of multimedia technology,more and more processors are integrated with SIMD（Single Instruction Multiple Data）extensions.Almost all current compilers are equipped with automatic vectorization.SLP（Superword Level Parallelism）method is introduced to some compilers in order to exploit intra-iteration parallelism.Frequent data type conversions are required when handling the multimedia data because of its characteristics of intensive storage and regular computation.However,the processing capacity of current SLP technique is not sufficient enough.In order to exploit more opportunities of SLP vectorization in programs with large amounts of type conversion statements,an SLP vectorization method was proposed to deal with type conversion statements.This method is able to handle the type conversion statements with the same or different vector factors using data regrouping in the framework of SLP vectorization.The experiment results show that the proposed method is efficient to exploit the SLP vectorization of data type conversion statements and is effective to improve the performance of the vectorization programs.

关 键 词：类型转换 数据重组 SLP SIMD 向量因子 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]