vector processing pdf

Page 1

CLICKHERETO DOWNLOAD

VectorFunctionalUnits(FUs):fullypipelined,startnewoperationeveryclockVectorProcessors(II)Avectorinstructionperformsanoperationoneachelement inconsecutivecycles.Inthispaper,wepresentAra2,thefirstfullyopen-sourcevectorprocessortosupporttheRISC-VVfrozenISAVectorprocessingoccurs whenarithmeticorlogicaloperationsareappliedtovectorsTheoperandtotheinstructionsarecompletevectorsinsteadofoneelementComponentsofVector ProcessorVectorprocessorsreducethefetchandodebandwidthasthenumberofinstructionsfetchedarelessDefinitionofasupercomputer:Fastestmachinein theworldatgiventaskTheconversionfromscalarcodetovectorcodeiscalledvectorizationNointra-vectordependenciesnohardwareinterlockingwithina vector1IntroductionVector()LectureVectorProcessingTointroducetheprinciplesofpipelining;Toillustratehowresourceconflictsarehandledinpipelined systems;TogivePropertiesofVectorProcessorsBothpipelinedprocessorsandSIMDcomputerscanperformvectoroperationsThisisincontrasttoascalar processorwhichhandlesoneelementatatimeusingmultipleinstructionsAbstract Vectorprocessingishighlyeffectiveinboostingprocessorperformanceand eficiencyfordata-parallelworkloads.AnymachinedesignedbySeymourCray.Vectorfunctionalunitsarepipelined.AvectorprocessorisaCPUdesignwherein theinstructionsetincludesoperationsthatcanperformmathematicaloperationsonmultipledataelementssimultaneouslyAVectorprocessorisaprocessorthat canoperateonanentirevectorinoneinstructionTheoperandtotheinstructionsarecompletevectorsinsteadofoneelementInChaptersandwesawhowwe couldsignificantlyincreasetheperformanceofaprocessorbyissuingmultipleinstructionsperclockcycleandbyAbstract Vectorprocessingishighlyeffectivein boostingprocessorperformanceandeficiencyfordata-parallelworkloadsAnymachinecosting$30M+KasimMAl-Aubidy–hasatleastreadandwriteports –typicallyvectorregisters,eachholdingbitelementsAdevicetoturnacompute-boundproblemintoanI/OboundproblemObjectivesComputerEng DeptVectorcomputersareequippedwithscalarandvectorhardwareorappearasSIMDAvectorprocessorisaCPUdesignwhereintheinstructionsetincludes operationsthatcanperformmathematicaloperationsonmultipledataelementssimultaneouslyItisdistinguishedfromscalarprocessingwhichoperatesononeor onepairofdataEachresultindependentofpreviousresultlongpipeline,compilerensuresnodependencieshighclockrateVectorinstructionsaccessmemorywith knownpatternhighlyinterleavedmemoryamortizememorylatencyofover≈elementsno(data)cachesrequired!VectorRegister:fixedlengthbankholdingasingle vectorThisisWhyVectorProcessors?Inthispaper,wepresentAra2,thefirstfullyopenVectorprocessorsarespecialpurposecomputersthatmatcharangeof (scientific)computingtasksCDC(Cray,)regardedasfirstsupercomputerVECTORPROCESSING&PARALLELPROCESSINGVectorProcessorsVector processorshavehigh-leveloperationsthatworkonlineararraysofnumbers:"vectors"eg,A=BxC,whereA,B,CareelementvectorsofAVectorprocessoris aprocessorthatcanoperateonanentirevectorinoneinstruction.Eachpipelinestageoperatesonadifferentdataelement.Vectorinstructionsallowdeeper pipelinesThesetasksusuallyconsistoflargeactivedatasets,oftenpoorChapterPipeliningandVectorProcessingProf

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.
vector processing pdf by 76l8iu2bxqj - Issuu