集成众核平台下CombLec88的并行化  

Parallelization of CombLec88 based on many integrated core

作　　者：宋博文[1,2] 张保东[1,2] 周晓辉[1,3,2] 

机构地区：[1]西安邮电大学计算机学院,陕西西安710121 [2]高效能服务器和存储技术国家重点实验室,山东济南250101 [3]陕西省高性能计算研究中心,陕西西安710121

出　　处：《计算机工程与设计》2016年第2期433-437,共5页Computer Engineering and Design

基　　金：高效能服务器和存储技术国家重点实验室开放基金项目(2014HSSA13);陕西省自然科学基础研究计划基金项目(2013JM8028)

摘　　要：针对随机数产生器CombLec88相对于传统的线性同余产生器LCG(linear congruential generator)运行速度较慢的情况,在研究CombLec88串行算法的基础上,利用"Simple skip ahead"算法并行化理论,提出一种基于Intel MIC(many integrated core)众核平台的CombLec88并行化方案。割分一个周期内的原始随机数序列,汇总各个线程产生的随机数,产生所需随机序列。实验结果表明,该方案能够通过TestU01的454项测试,与串行程序相同,移植到Intel MIC平台后性能良好,相对CPU单线程的最优加速比为16.004。Random number generator CombLec88＇s speed is lower than conventional random number generator LCG（linear congruential generator）.The issue that how to implement the parallelization of CombLec88 based on many integrated core（MIC）was discussed,taking advantage of the parallelization of algorithm,by means of partitioning the original random number sequence into different subsequences,the random number gererated by all threads were combined into the final sequence.Experimental results show that the parallelized CombLec88 generator successfully passes 454 tests of TestU01,the results are the same as CombLec88 generator without parallelization.The speedup based on the MIC is very impressive,and the best speedup reaches16.004 relative to single-thread with the CPU.

关 键 词：随机数产生器 并行化 CombLec88 集成众核 TestU01 

分 类 号：TP311.52[自动化与计算机技术—计算机软件与理论]