Energy efficient architecture research has flourished recently, in an attempt to address packaging and cooling concerns of current microprocessor designs, as well as battery life for mobile computers. Moreover, architects have become increasingly concerned
InProceedingsofthe4thInternationalSymposiumonHighPerformanceComputing(ISHPC),May2002,(c)Springer-Verlag.HighPerformanceandEnergyEf cientSerialPrefetchArchitecture
GlennReinmanBradCalderToddAustin
ComputerScienceDepartment,UniversityofCalifornia,LosAngeles
DepartmentofComputerScienceandEngineering,UniversityofCalifornia,SanDiego
ElectricalEngineeringandComputerScienceDepartment,UniversityofMichigan
Abstract
Energyef cientarchitectureresearchhas ourishedrecently,inanattempttoaddresspackaging
andcoolingconcernsofcurrentmicroprocessordesigns,aswellasbatterylifeformobilecomputers.
Moreover,architectshavebecomeincreasinglyconcernedwiththecomplexityoftheirdesignsinthe
faceofscalability,veri cation,andmanufacturingconcerns.
Inthispaper,weproposeandevaluateahighperformance,energyandcomplexityef cientfront-end
prefetcharchitecture.Thisdesign,calledSerialPrefetching,combinesahighfetchbandwidthbranch
predictionandef cientinstructionprefetchingarchitecturewithalow-energyinstructioncache.Serial
Prefetchingexploresthebene tofdecouplingthetagcomponentofthecachefromthedatacomponent.
Cacheblocksare rstveri edbythetagcomponentofthecache,andthentheaccessesareputinto
aqueuetobeconsumedbythedatacomponentoftheinstructioncache.Energyissavedbyonly
accessingthecorrectwayofthedatacomponentspeci edbythetaglookupinapreviouscycle.The
tagcomponentdoesnotstallonaI-cachemiss,onlythedatacomponent.Theaccessesthatmissinthe
tagcomponentarespeculativelybroughtinfromlowerlevelsofthememoryhierarchy.Thisineffect
performsaprefetch,whiletheaccessmigratesthroughthequeuetobeconsumedbythedatacomponent.
1Introduction
Atahigh-level,amodernhigh-performanceprocessoriscomposedoftwoprocessingengines:thefront-endprocessorandtheexecutioncore.Thisproducerandconsumerrelationshipbetweenthefront-endandexecutioncorecreatesafundamentalbottleneckincomputing,i.e.,executionperformanceisstrictlylimitedbyfetchperformance.
Anenergyef cientfetchdesignthatstillachieveshighperformanceisimportantbecauseoverallchipenergyconsumptionmaylimitnotonlywhatcanbeintegratedontoachip,butalsohowfastthechipcanbeclocked[6].Brooksetal.[2]reportthatinstructionfetchandthebranchtargetbufferareresponsiblefor22.2%and4.7%respectivelyofpowerconsumedbytheIntelPentiumPro.Brooksalsoreportsthatcachescomprise16.1%ofthepowerconsumedbyAlpha21264.Montanaroetal.[5]foundthattheinstructioncacheconsumes27%ofpowerintheirStrongARM110processor.
Thegoaloftheresearchpresentedinthispaperistocreateafetchenginethatprovides:
1.highfetchbandwidthforwideissueprocessors
2.acomplexityeffectivedesignthatwillscaletofutureprocessortechnologies
1
Sixteen levels (double-words) of prefetch buffers from DRAM for access by ...high performance, cost-effective and energy efficient chip set for the ...
Dissertation Enabling Efficient High-Performance Communication In_专业资料。To teach courses in computer architecture, computer engineering, and VLSI design, ...
cheaper, more energy-efficient and more reliable... prefetch few instructions – 8088 (8 bit ... – High-end servers – 64-bit architecture ? ...
architecture with separate local instruction and ... and includes an internal prefetch unit that ...performance with a simple instruction set and ...
Journal of Systems Architecture Aerospace Science ...and developing the most efficient digit-serial DSP...high-performance multi-channel counter Original ...
An Efficient Architecture for Loop Based Data ...Sicolo Center for Reliable and High-Performance ...This special prefetch instruction informs the memory...
Efficient Digit-Serial Rational Function Approximations and Digital Filtering ...lter performance. Given a target transfer function ? (for low pass ?lter)...
Cache High Performance N... 暂无评价 38页 ...of IP core, a prefetch technique is presented. ...In the architecture, each IP core may (or may...
Cache High Performance N... 暂无评价 38页 ...of IP core, a prefetch technique is presented. ...In the architecture, each IP core may (or may...
热门文档