2.1,2 CPU (中央处理器)

Chapter2CPUIntroductiontoCPUCentralprocessingunitetchedonsiliconchipcalledmicroprocessorContaintensofmillionsoftinytransistorsKeycomponents:CentralprocessingunitRegistersSystemclockTypesofChipsIntelmakesafamilyofprocessorsPentiumIIIandPentium4processorsinmostPCsCeleronprocessorsoldforlow-costPCsXeonandItaniumforhigh-endworkstationsandnetworkserversOtherprocessorsCyrixandAMDmakeIntel-compatiblemicroprocessorsPowerPCchipsusedprimarilyinMacintoshcomputersHP’sAlphamicroprocessorusedinhigh-endserversMicroprocessorSpeedsMeasureofsystemclockspeedHowmanyelectronicpulsestheclockproducespersecondUsuallyexpressedingigahertz(GHz)BillionsofmachinecyclespersecondSomeoldPCsmeasuredinmegahertz(MHz)ComparisonofclockspeedonlymeaningfulbetweenidenticalmicroprocessorsCPUcycletime–inverseofclockrateCurrentTechnologyCapabilitiesandLimitationsMoore’sLawRateofincreaseintransistordensityonmicrochipsdoublesevery18-24monthswithnoincreaseinunitcostRock’sLawCostoffabricationfacilitiesforchipgenerationdoubleseveryfouryearsIncreasedpackingdensityElectricalresistance3.2ComponentsoftheCPUControlunitMovesdataandinstructionsbetweenmainmemoryandregistersArithmeticlogicunit(ALU)PerformscomputationandcomparisonoperationsSetofregistersStoragelocationsthatholdinputsandoutputsfortheALUActionsPerformedbyCPUFetchcycleCPU:FetchesaninstructionfromprimarystorageIncrementsapointertolocationofnextinstructionSeparatesinstructionintocomponents(instructioncodeanddatainputs)StoreseachcomponentinaseparateregisterExecutioncycleALU:RetrievesinstructioncodefromaregisterRetrievesdatainputsfromregistersPassesdatainputsthroughinternalcircuitstoperformdatatransformationStoresresultsinaregisterCPURegistersPrimaryrolesHolddataforcurrentlyexecutingprogramthatisneededquicklyorfrequently(general-purposeregisters)StoreinformationaboutcurrentlyexecutingprogramandaboutstatusofCPU(special-purposeregisters)General-PurposeRegistersHoldintermediateresultsandfrequentlyneededdataitemsUsedonlybycurrentlyexecutingprogramImplementedwithintheCPU;contentscanbereadorwrittenquicklyIncreasingtheirnumberusuallydecreasesprogramexecutiontimetoapointSpecial-PurposeRegistersTrackprocessorandprogramstatusTypesInstructionregisterInstructionpointerProgramstatusword(PSW)StoresresultsofcomparisonoperationControlsconditionalbranchexecutionIndicatesactualorpotentialerrorconditionsWordSizeNumberofbitsaCPUcanprocesssimultaneouslyIncreasingitusuallyincreasesCPUefficiency,uptoapointOthercomputercomponentsshouldmatchorexceeditforoptimalperformanceImplicationsforsystembusdesignandphysicalimplementationofmemory3.3ThePhysicalCPUElectricaldeviceimplementedassilicon-basedmicroprocessorContainsmillionsofswitches,whichperformbasicprocessingfunctionsPhysicalimplementationofswitchesandcircuitsTransistorsElectronicswitchesthatmayormaynotallowelectriccurrenttopassthroughIfcurrentpassesthrough,switchison,representinga1bitOtherwise,switchisoff,representinga0bitSwitchesandGatesBasicbuildingblocksofcomputerprocessingcircuitsElectronicswitchesControlelectricalcurrentflowinacircuitImplementedastransistorsGatesAninterconnectionofswitchesAcircuitthatcanperformaprocessingfunctiononanindividualbinaryelectricalsignal,orbitElectricalPropertiesConductivityAbilityofanelementtoenableelectronflowResistanceLossofelectricalpowerthatoccurswithinaconductorHeatNegativeeffectsofheat:PhysicaldamagetoconductorChangestoinherentresistanceofconductorDissipateheatwithaheatsinkSpeedandcircuitlengthTimerequiredtoperformaprocessingoperationisafunctionoflengthofcircuitandspeedoflightReducecircuitlengthforfasterprocessingProcessorFabricationPerformanceandreliabilityofprocessorshasincreasedwithimprovementsinmaterialsandfabricationtechniquesTransistorsandintegratedcircuits(ICs)MicrochipsandmicroprocessorsFirstmicroprocessor(1971)–2,300transistorCurrentmemorychip–300milliontransistors3.4FutureTrendsSemiconductorsareapproachingfundamentalphysicalsizelimitsTechnologiesthatmayimproveperformancebeyondsemiconductorlimitationsOpticalprocessingHybridoptical-electricalprocessingQuantumprocessingOpticalProcessingCouldeliminateinterconnectionandsimplifyfabricationproblems;photonpathwayscancrosswithoutinterferingwithoneanotherEliminatingwireswouldimprovefabricationcostandreliabilityNotenougheconomicincentivetobearealityyetElectro-OpticalProcessingDevicesprovideinterfacebetweensemiconductorandpurelyopticalmemoryandstoragedevicesGalliumarsenide(bothopticalandelectricalproperties)Silicon-basedsemiconductordevices(encodedatainexternallygeneratedlaserlight)QuantumProcessingUsesquantumstatestosimultaneouslyencodetwovaluesperbit(qubit)UsesquantumprocessingdevicestoperformcomputationsTheoreticallywell-suitedtosolvingproblemsthatrequiremassiveamountsofcomputation