专业英语论文翻译[精选].docx

专业英语论文翻译[精选]

第一篇：专业英语论文翻译[精选]AParallelizationCostModelforGPUGPU并行成本模型2009137127周幼兰Abstract-UsingGPUforgeneralcomputinghasbecomeanimportantresearchdirectioninhighperformancecomputingtechnology.However,thisisnotalosslessoptimizationmethod.Duetotheimpactofdeviceinitializationcost,datatransmissiondelay,specificcharacteristicsofprograms,andotherfactors,thegeneralcomputingonGPUmaynotalwaysachievethedesiredspeedup,andsometimesresultsinprogramexecutionperformancedegradation.Onthebasisofin-depthanalysisofGPUinternalprocessingmechanisms,themainfactorsaffectingGPUimplementationperformancearepointedout,andaparallelcostmodelforGPUbasedonstaticprogramanalysisisproposedtoprovidejudgementbasisforusingGPUingeneralcomputing.摘要：在高性能计算技术领域，使用GPU执行通用计算已成为一个重要的研究方向。但它并不是一种无损优化方法。由于受设备初始化成本、数据传输延迟、程序本身特征和一些其他因素的影响，基于GUP的通用计算不可能总是达到预期的加速，有时还会导致程序执行性能下降。在对GPU内部处理机制进行深度分析的基础上，得到影响GPU执行性能的主要因素，且得出基于静态程序分析的GPU并行成本模型为GPU在通用计算中的使用提供了判断依据的结论。I.INTRODUCTIONGraphicsprocessingUnit(GPU)hasdevelopedataspeedmuchfasterthantheMoore’sLawinrecentyears,notonlyimprovingimageprocessing,virtualreality,computersimulation,andthedevelopmentofrelatedapplications,butalsoprovidingagoodrunningplatformforgeneral-purposecomputingusingGPUbeyondgraphicprocessing.TheapplicationofGPUingeneral-purposecomputingmakesaseriesofnewchallengesfacedbythedevelopmentofthehighperformancecomputingtechnology.Currently,inordertoreducetheprogrammingcomplexityofGPUingeneral-purposecomputing,manyGPUmanufacturersandresearchinstitutionsproposedanumberofprogramminglanguagesandprogrammingmodelsclosetotraditionalprogrammingmethods,butdifferentstyles,suchasBrook+[1],CUDA[2]andOpenCL[3],etc.However,asGPUhasitsownspecificcharacteristicofhardwarearchitectureanddevelopment,programmersmusthaveahighlevelofexpertise.UsingGPUforgeneral-purposecomputing,theincreasedperformanceachievedinlargepartdependsonthehardwareknowledgeandprogrammingskillsofprogrammers.Atpresent,themoststudiesfortheGPU’sparallelismbothathomeandabroaddirectlyrewriteandtransplantprogra