关键词: 类型转换,数据重组,SLP,SIMD,向量因子

Abstract: 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 paralle-lism.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.

Key words: Type conversion,Data regrouping,SLP,SIMD,Vector factor