摘要：
Firstly, we introduced the single Instruction Multiple Data(SIMD) vectorization technology and the features separately,based on the processors of Intel AMD and IBM CelL Secondly, some vectorization functions were tested in these three platforms, which were deve-loped by the three vendors separately. Our test results show that we achieve high performance with the technology of the vectorization,compared to the traditional methods of the scalar calculation. Especially, the speedup of the Cell SDK functions is 10 on average, which were achieved by the help of many processing elements and the special processor structure. Lastly, we also found that there are some differences between the vectorial functions, which are in different vector math libraries. We analyzed that there are some reasons caused the difference between the math functions in different platforms,such as processor structure, the number of the processing elements,memery accessing and so on.%首先,结合Intel,AMD和IBM处理器,介绍了单指令流多数据流(SIMD)向量化技术及其各自的特点.其次,在3种平台上对各自开发的函数库中的部分向量数学函数进行了测试.结果表明,相对传统的标量计算,向量化技术带来的加速比较高,特别是Cell SDK函数,因其独特的体系结构,多个向量处理单元带来的平均加速比为10.最后,通过测试结果的对比,发现不同数学库中的向量函数之间在性能方面也存在着差异,并对差异原因进行了分析,得出性能差异主要是处理器架构和向量计算单元个数和访存等因素造成的.
