A Hybrid Circular Queue Method for Iterative Stencil Computations on GPUs-在gpu上迭代模板计算的混合循环队列方法.pdfVIP

for on XBeta1．A circularmethoditerativestencil YangY，CutHM，Feng hybrid queue computations OFCOMPUTERSCIENCEANDTECHNOLOGYJan．2012．DOI 27f1)：57—74 10．i007／s11390—012—1206—3 A CircularMethodforIterativeStencil Hybrid Queue onGPUs Fen91(冯晓兵)，Member，CCF YangYan91，2(杨杨)，Hut—MinCutl，2(崔慧敏)，Xiao—Bing and Member，IEEE Xue3(薛京灵)，Senior Jing-Ling 1State AcademyolSciences LaboratoryofComputerArchitecture，[nst{tuteo|ComputingTechnology}Chinese Key Beijing100190，China 2Graduate Chinese 1001 of AcademyolSciencesjBeijing90,China University and of Scienceand 3ProgrammingLan9uagesCompilersGroup，SchoolComputer Engineering NewSouth 2052，Australia Universityof Wales．勖dney,NSW E—math{yangyang，cuihm，fxb},．@ict．ac．on；jingling@cse．unsw．edu．au Received October May9，2011；revised8，2011 a thatcan boostthe ofstencil AbstractInthis circularmethod performance paper，wepresenthybrid queue significantly of and earliermethodsthat on onGPU rely computationsbycarefullybalancingusageregistersshared—memory．Unlike addressableshared method anew circular hybrid exploits queuespredominantlyitnplementedusingindirecoy lnentory，our acrossthe time instencil sothatcircular canbe