【文档说明】计算机组成原理英文版第一章Computer-Abstractions-and-Technology--课件.ppt，共(63)页，8.526 MB，由小橙橙上传

转载请保留链接：https://www.ichengzhen.cn/view-76280.html

以下为本文档部分文字说明：

Chapter1ComputerAbstractionsandTechnologyTheComputerRevolutionProgressincomputertechnologyUnderpinnedbyMoore

’sLawWhatisMoore’sLaw???Moore'slawdescribesalong-termtrendinthehistoryofcomputinghardware.Thequantityoftransistorstha

tcanbeplacedinexpensivelyonanintegratedcircuithasdoubledapproximatelyeverytwoyears.§1.1Introductionppt课件Moore’sLawThetrendhascontin

uedformorethanhalfacenturyandisnotexpectedtostopuntil2015orlater.ppt课件TheComputerRevolutionMakesnovelap

plicationsfeasibleComputersinautomobiles§1.1Introductionppt课件TheComputerRevolutionMakesnovelapplicationsfeasibleCellphones§1.1Introductionppt课件T

heComputerRevolutionMakesnovelapplicationsfeasibleHumangenomeprojectWorldWideWebSearchEngines§1.1Introductionppt课件T

heComputerRevolutionComputersarepervasive§1.1Introductionppt课件ClassesofComputersQuestion:Howdoyouclassifycomp

uters?DesktopComputersServerComputersEmbeddedComputersppt课件ClassesofComputersDesktopcomputersPC

Generalpurpose,varietyofsoftwareSubjecttocost/performancetradeoffppt课件ClassesofComputersServercomputersNetworkbasedHighcapacit

y,performance,reliabilityRangefromsmallserverstobuildingsizedWorld’ssmallestwebserverppt课件ClassesofComputersEmbeddedcomputersHiddenascompo

nentsofsystemsStringentpower/performance/costconstraintsppt课件TheProcessorMarketppt课件WhatYouWillLearnHo

wprogramsaretranslatedintothemachinelanguageAndhowthehardwareexecutesthemThehardware/softwareinterfaceWhatdeterminesprogramperformanceAndho

witcanbeimprovedHowhardwaredesignersimproveperformanceWhatisparallelprocessingppt课件LevelsofProgramCodeHigh-l

evellanguageLevelofabstractionclosertoproblemdomainProvidesforproductivityandportabilityAssemblylan

guageTextualrepresentationofinstructionsHardwarerepresentationBinarydigits(bits)Encodedinstructionsa

nddatappt课件BelowYourProgramApplicationsoftwareWritteninhigh-levellanguage(HLL)SystemsoftwareCompiler:tran

slatesHLLcodetomachinecodeOperatingSystem:servicecodeHandlinginput/outputManagingmemoryandstorageSchedulingtasks&sharingr

esourcesHardwareProcessor,memory,I/Ocontrollers§1.2BelowYourProgramppt课件UnderstandingPerformanceAlgorithmDeterminesnumberofoperati

onsexecutedProgramminglanguage,compiler,architectureDeterminenumberofmachineinstructionsexecutedperoperationP

rocessorandmemorysystemDeterminehowfastinstructionsareexecutedI/Osystem(includingOS)DetermineshowfastI/Ooperationsareexecut

edppt课件ComponentsofaComputerSamecomponentsforallkindsofcomputerDesktop,server,embeddedInput/outputincludesUser-in

terfacedevicesDisplay,keyboard,mouseStoragedevicesHarddisk,CD/DVD,flashNetworkadaptersForcommunicatingwithothercomputers§

1.3UndertheCoversTheBIGPictureppt课件Anatomy(结构)ofaComputerOutputdeviceInputdeviceInputdeviceNetworkcableppt课件AnatomyofaMouseO

pticalmouseLEDilluminatesdesktopSmalllow-rescameraBasicimageprocessorLooksforx,ymovementButtons&wheelSupersedesroller-ballmechanicalmous

eppt课件ThroughtheLookingGlassLCDscreen:pictureelements(pixels)Mirrorscontentofframebuffermemoryppt课件Openin

gtheBoxppt课件InsidetheProcessor(CPU)Datapath:performsoperationsondataControl:sequencesdatapath,memory,...CachememorySmallfastSRAMmemoryfori

mmediateaccesstodataSRAM–StaticRandomAccessMemoryppt课件InsidetheProcessorAMDBarcelona:4processorcoresppt课件ASafePlaceforData

Volatile(易变的)mainmemoryLosesinstructionsanddatawhenpoweroffNon-volatilesecondarymemoryMagneticdiskFlashmemoryOpticaldisk(CDROM,DVD)pp

t课件NetworksCommunicationandresourcesharingLocalareanetwork(LAN):EthernetWithinabuildingWideareanetwork(WAN):theInternet

ppt课件AbstractionsInstructionSetArchitecture(ISA)Aninterfacebetweenthehardwareandthelowest-levelsoftwareTheabstractimageofacomputing

systemthatisseenbyamachine/assemblylanguageprogrammerIncludinginstructions,registers,memoryaccess,I/O,…TheBIGPictu

reppt课件ISAsSystem/360andupwardscompatiblesuccessorsz/ArchitecturePowerArchitecturePDP-11SPARCSuperHTricoreTransputerUNIVAC1100/2200series

VAXx86IA-32(32-bitx86,firstimplementedintheIntel80386)x86-64(64-bitsupersetofIA-32,firstimplementedintheAM

DOpteron)EISC(AE32K)4004,40406800,6502,6809,68HC11,68HC08.8008,8080,8085,Z80,Z180,eZ80,etc.8048,8051,etc.

Z8,eZ8,etc.AlphaARMBurroughsB5000seriesBurroughsB6000/B7000serieseSi-RISCIA-64(Itanium)Mico32MIPS

Motorola68kPA-RISCIBM700/7000linesppt课件AbstractionsApplicationBinaryInterface(ABI)thelow-levelinterfacebetweenanapplicationprogram

andOSABIscoverdetailssuchasdatatype,size,alignment,callingconvention,binaryformatofobjectfiles,etc.Definesastandardforbinaryportability

acrosscomputers.ppt课件AbstractionsImplementationHardwarethatobeysthearchitectureabstractionManyimplementationsforthesameISAExample:IntelPenti

umvsAMDAthlon,almostidenticalISA,butradicallydifferentinternaldesigns.ppt课件PERFORMANCEppt课件DefiningPerformanceWhichairplanehasthebestperformance?§1

.4Performanceppt课件DefiningPerformanceWhichairplanehasthebestperformance?0100200300400500DouglasDC-8-50BAC/SudConcordeBoeing747Boeing777Passe

ngerCapacity0200040006000800010000DouglasDC-8-50BAC/SudConcordeBoeing747Boeing777CruisingRange(miles)050010001500DouglasDC-8-5

0BAC/SudConcordeBoeing747Boeing777CruisingSpeed(mph)0100000200000300000400000DouglasDC-8-50BAC/SudConcordeBoeing747Boeing777Passengers

xmph§1.4Performanceppt课件DefiningPerformancePerformance=SpeedE.g.,takingasinglepassengerfromonepointtoanotherWinner

:ConcordePerformance=PassengerThroughputPassengersxm.p.hE.g.,transporting450passengersfromonepointtoanotherWinner:B

oeing747ppt课件ResponseTimeandThroughputResponsetimeHowlongittakestodoataskThroughputTotalworkdoneperunittimee.g.,ta

sks/transactions/…perhourppt课件ResponseTimeandThroughputHowareresponsetimeandthroughputaffectedbyReplacing

theprocessorwithafasterversion?Addingmoreprocessors?We’llfocusonresponsetimefornow…ppt课件RelativePerformanceDefi

nePerformance=1/ExecutionTime“XisntimefasterthanY”nXYYXtimeExecutiontimeExecutionePerformancePerformancExample:timetakentor

unaprogram10sonA,15sonBExecutionTimeB/ExecutionTimeA=15s/10s=1.5SoAis1.5timesfasterthanBppt课件MeasuringExecutionTime

ElapsedtimeTotalresponsetime,includingallaspectsProcessing,I/O,OSoverhead,idletimeDeterminessystemperformanceCPUti

meTimespentprocessingagivenjobDiscountsI/Otime,otherjobs’sharesComprisesuserCPUtimeandsystemCPUtimeDifferentprogramsareaffecteddifferentlybyCPUa

ndsystemperformanceppt课件CPUClockingOperationofdigitalhardwaregovernedbyaconstant-rateclockClock(cycles)Datatransferandco

mputationUpdatestateClockperiodClockcycletime:durationofaclockcyclee.g.,250ps=0.25ns=250×10–12sClockfrequency(clockrate):cyclespersecond

e.g.,4.0GHz=4000MHz=4.0×109Hzppt课件CPUTimePerformanceimprovedbyReducingnumberofclockcyclesIncreasing

clockrateHardwaredesignermustoftentradeoffclockrateagainstcyclecountRateClockCyclesClockCPUTimeCycleClockCyclesC

lockCPUTimeCPUppt课件CPUTimeExampleComputerA:2GHzclock,10sCPUtimeDesigningComputerBAimfor6sCPUtimeCandofasterclock,butcauses1.2×cloc

kcyclesHowfastmustComputerB’sclockratebe?4GHz6s10246s10201.2RateClock10202GHz10sRateClockTimeCPUCyclesC

lock6sCyclesClock1.2TimeCPUCyclesClockRateClock99B9AAAABBBppt课件InstructionCountandCPIInstructionCountforaprogra

mDeterminedbyprogram,ISAandcompilerAveragecyclesperinstructionDeterminedbyCPUhardwareIfdifferentin

structionshavedifferentCPIAverageCPIaffectedbyinstructionmixRateClockCPICountnInstructioTimeCycleClockCPICountnInstructioTimeCPUnInstructioperCycles

CountnInstructioCyclesClockCPIppt课件CPIExampleComputerA:CycleTime=250ps,CPI=2.0ComputerB:CycleTime=500ps,CPI=1.2SameIS

AWhichisfaster,andbyhowmuch?1.2500psI600psIATimeCPUBTimeCPU600psI500ps1.2IBTimeCycleBCPICountnInstr

uctioBTimeCPU500psI250ps2.0IATimeCycleACPICountnInstructioATimeCPUAisfaster……bythismuchppt课件CPIinMoreDetail

Ifdifferentinstructionclassestakedifferentnumbersofcyclesn1iii)CountnInstructio(CPICyclesClockWeightedaverageCPI

n1iiiCountnInstructioCountnInstructioCPICountnInstructioCyclesClockCPIRelativefrequencyppt课件CPIExampleAlternativecompiledcodesequencesusinginstr

uctionsinclassesA,B,CClassABCCPIforclass123ICinsequence1212ICinsequence2411Sequence1:IC=5ClockCycles=2×1+1×2+2×3=10Av

g.CPI=10/5=2.0Sequence2:IC=6ClockCycles=4×1+1×2+1×3=9Avg.CPI=9/6=1.5Whatisavg.CPI?IC=InstructionCountppt课件PerformanceSummaryPerformancedependson

Algorithm:affectsIC,possiblyCPIProgramminglanguage:affectsIC,CPICompiler:affectsIC,CPIInstructionsetarchitec

ture:affectsIC,CPI,TcTheBIGPicturecycleClockSecondsnInstructiocyclesClockProgramnsInstructioTimeCPUppt课件POWERppt课件PowerT

rendsInCMOSICtechnology§1.5ThePowerWallFrequencyVoltageloadCapacitivePower2×1000×305V→1Vppt课件ReducingPowerSupposeanewCPUhas85%ofca

pacitiveloadofoldCPU15%voltageand15%frequencyreduction0.520.85FVC0.85F0.85)(V0.85CPP4old2oldoldold2oldoldoldnewThepowerwall

Wecan’treducevoltagefurtherWecan’tremovemoreheatHowelsecanweimproveperformance?ppt课件UniprocessorPerformance§1.6

TheSeaChange:TheSwitchtoMultiprocessorsConstrainedbypower,instruction-levelparallelism,memorylatencypp

t课件MultiprocessorsMulticoremicroprocessorsMorethanoneprocessorperchipRequiresexplicitlyparallelprogrammingCompar

ewithinstructionlevelparallelismHardwareexecutesmultipleinstructionsatonceHiddenfromtheprogrammerHardtodoProgramming

forperformanceLoadbalancingOptimizingcommunicationandsynchronizationppt课件MANUFACTURINGppt课件Manufacturi

ngICsYield:proportionofworkingdiesperwafer§1.7RealStuff:TheAMDOpteronX4ppt课件AMDOpteronX2WaferX2:300mmwafer,117chips,90nmtechnologyX4

:45nmtechnologyppt课件IntegratedCircuitCostNonlinearrelationtoareaanddefectrateWafercostandareaarefixedDefectratedeterminedbymanufacturingproce

ssDieareadeterminedbyarchitectureandcircuitdesign2area/2))Dieareaper(Defects(11YieldareaDieareaWaferwaferp

erDiesYieldwaferperDieswaferperCostdieperCostppt课件BENCHMARKINGppt课件SPECCPUBenchmarkProgramsusedtomeasureperfo

rmanceSupposedlytypicalofactualworkloadStandardPerformanceEvaluationCorp(SPEC)DevelopsbenchmarksforCPU,I/O,Web,…SPECCPU2006Elapsedti

metoexecuteaselectionofprogramsNegligibleI/O,sofocusesonCPUperformanceNormalizerelativetoreferencemachineSummarizeasgeometricmeanofperform

anceratiosCINT2006(integer)andCFP2006(floating-point)nn1iiratiotimeExecutionppt课件CINT2006forOpteronX42356NameDescriptionIC×109CPITc(

ns)ExectimeReftimeSPECratioperlInterpretedstringprocessing2,1180.750.406379,77715.3bzip2Block-sortingcompression2,3890.850.408179,65011.8gccG

NUCCompiler1,0501.720.47248,05011.1mcfCombinatorialoptimization33610.000.401,3459,1206.8goGogame(AI)1,6581.090.40

72110,49014.6hmmerSearchgenesequence2,7830.800.408909,33010.5sjengChessgame(AI)2,1760.960.483712,10014.5libquantumQuantumcomp

utersimulation1,6231.610.401,04720,72019.8h264avcVideocompression3,1020.800.4099322,13022.3omnetppDiscr

eteeventsimulation5872.940.406906,2509.1astarGames/pathfinding1,0821.790.407737,0209.1xalancbmkXMLparsing1,0582.700.401,1

436,9006.0Geometricmean11.7ppt课件SPECPowerBenchmarkPowerconsumptionofserveratdifferentworkloadlevelsPerformance:ssj_ops/secPower:Watts(Joules/sec)

100ii100iipowerssj_opsWattperssj_opsOverallppt课件SPECpower_ssj2008forX4TargetLoad%Performance(ssj_o

ps/sec)AveragePower(Watts)100%231,86729590%211,28228680%185,80327570%163,42726560%140,16025650%118,32424640%920,3523330%70,50022220%47,12

620610%23,0661800%0141Overallsum1,283,5902,605∑ssj_ops/∑power493ppt课件Pitfall:Amdahl’sLawImprovinganaspectofacomputerandexpectingaproportionalim

provementinoverallperformance§1.8FallaciesandPitfalls208020nCan’tbedone!unaffectedaffectedimprovedTfactortimprovemenTTExample:multiplyaccounts

for80s/100sHowmuchimprovementinmultiplyperformancetoget5×overall?ppt课件Fallacy:LowPoweratIdleLookbackatX4powe

rbenchmarkAt100%load:295WAt50%load:246W(83%)At10%load:180W(61%)GoogledatacenterMostlyoperatesat10%–50%loadAt100%loadlessthan1%ofth

etimeConsiderdesigningprocessorstomakepowerproportionaltoloadppt课件Pitfall:MIPSasaPerformanceMetricMIPS:MillionsofInstructionsPerS

econdDoesn’taccountforDifferencesinISAsbetweencomputersDifferencesincomplexitybetweeninstructions66610CPIrateClock10rateClockCPIcountnIns

tructiocountnInstructio10timeExecutioncountnInstructioMIPSCPIvariesbetweenprogramsonagivenCPUppt

课件ConcludingRemarksCost/performanceisimprovingDuetounderlyingtechnologydevelopmentHierarchicallayersofabstractionInbothhardwa

reandsoftwareInstructionsetarchitectureThehardware/softwareinterfaceExecutiontime:thebestperformancemeasurePowerisalimitingfactorUsepa

rallelismtoimproveperformance§1.9ConcludingRemarksppt课件