基于动态profiling技术的流水粒度调优.pdfVIP

计算机研究与发展 IssN 1000—1239／CN 11—1777／TP Joumal of Computer ReSearch and Development 42(6)：1065～1072，2005 基于动态profiling技术的流水粒度调优 马 琳1’2 陈 莉1 冯晓兵1 1(中国科学院计算技术研究所北京100080) 2(中国科学院研究生院北京 100039) (malin@ict．ac．cn) Tuning Pipeline Granularity Based on Dynamic Profiling Framework Ma Linl～，Chen Lil，and Feng Xiaobin91 1(J扎sfif“匏盯CD优加fi咒g n咖咒ozq鲥，吼i托聊Ac口如my 0，i已咒唧，＆巧i729 100080) 2(Gmd“口抬Sc危00Z盯琥e吼i咒伽A∞如仍，o，iP，z唧，聊i咒g 100039) Abstract Pipelining is one of useful parallelization techniques for those loops which haVe cross-procesSor data dependences． And the pipeline granularity is the key to make the computation time be suitable for communication time and obtain g∞d pipeline perfomance．Loop strip—mining and loop interchange are g∞d methods to help find the optimal pipeline granularity． And the amount of computation between communication operations in each node is caUed pipeline granularity or block size．A 10t of factors decide the optimal pipeline granularity， such as access mode of application program， program size， total computing node，computation ability and memory architecture of the computing node，perfomance of communication network， communication mode， and overheads of runtime library， etc． It’s hard to assume the block computation time by using static scheme，and the run time scheme will have more extra runtime oVerhead and may 10se more optimization o{the application．An approach is presented and realized to compute the pipeline granularity by dynamic profiling and the cost model including the cache 10cality by lo。p transform． How to decrease the time of profiling running and guarantee the precision of the cost model is also considered．The results of the experiments prove that the pipeline granularity achieVed by dynamic profiling framework has good adaptability and speedup of the execution time of pipelined l∞p． Key words pipelined l∞p；dynamic profiling；l∞p transform 摘要结点间流水是解决数据分布和计算分割不一致时的一种重要的并行发掘技术．结点间流水通 过计算与通信的重叠获得并行度．精确的流水粒度是获得良好的流水性能的关键．流水分块取决于很 多因素，如程序规模、程序的访问模式、结