計算機組織與結(jié)構：版7-13-RISC

WilliamStallings

ComputerOrganization

andArchitecture

7thEditionChapter13ReducedInstructionSetComputers1Chapter13

ReducedInstructionSetComputersKeytermsCISCcomplexinstructionsetcomputerRISCreducedinstructionsetcomputerDelayedbranchDelayedloadHLLhigh-levellanguageRegisterfileRegisterwindowSPARC2MajorAdvancesinComputers(1)ThefamilyconceptIBMSystem/3601964DECPDP-8SeparatesarchitecturefromimplementationMicroporgrammedcontrolunitIdeabyWilkes1951ProducedbyIBMS/3601964CachememoryIBMS/360model8519693MajorAdvancesinComputers(2)SolidStateRAM(Seememorynotes)MicroprocessorsIntel40041971PipeliningIntroducesparallelismintofetchexecutecycleMultipleprocessors4TheNextStep-RISCReducedInstructionSetComputerKeyfeaturesLargenumberofgeneralpurposeregistersoruseofcompilertechnologytooptimizeregisteruseLimitedandsimpleinstructionsetEmphasisonoptimisingtheinstructionpipeline5Comparisonofprocessors6DrivingforceforCISCSoftwarecostsfarexceedhardwarecostsIncreasinglycomplexhighlevellanguagesSemanticgapLeadsto:LargeinstructionsetsMoreaddressingmodesHardwareimplementationsofHLLstatementse.g.CASE(switch)onVAXSemantic:語義的;語義學的

Gap英音：[g?p]豁口,裂口

7IntentionofCISCEasecompilerwritingImproveexecutionefficiencyComplexoperationsinmicrocodeSupportmorecomplexHLLsease減輕

HLL

縮寫詞abbr.

high-levellanguage【電腦】高級語言8ExecutionCharacteristicsOperationsperformedOperandsusedExecutionsequencingStudieshavebeendonebasedonprogramswritteninHLLsDynamicstudiesaremeasuredduringtheexecutionoftheprogram9OperationsAssignmentsMovementofdataConditionalstatements(IF,LOOP)SequencecontrolProcedurecall-returnisverytimeconsumingSomeHLLinstructionleadtomanymachinecodeoperationsAssignment

分配;指派,選派10WeightedRelativeDynamicFrequencyofHLLOperations[PATT82a]

DynamicOccurrenceMachine-InstructionWeightedMemory-ReferenceWeighted

PascalCPascalCPascalCASSIGN45%38%13%13%14%15%LOOP5%3%42%32%33%26%CALL15%12%31%33%44%45%IF29%43%11%21%7%13%GOTO—3%————OTHER6%1%3%1%2%1%11OperandsMainlylocalscalarvariablesOptimisationshouldconcentrateonaccessinglocalvariables

PascalCAverageIntegerConstant16%23%20%ScalarVariable58%53%55%Array/Structure26%24%25%12ProcedureCallsVerytimeconsumingDependsonnumberofparameterspassedDependsonlevelofnestingMostprogramsdonotdoalotofcallsfollowedbylotsofreturnsMostvariablesarelocal(c.f.localityofreference)13ImplicationsBestsupportisgivenbyoptimisingmostusedandmosttimeconsumingfeaturesLargenumberofregistersOperandreferencingCarefuldesignofpipelinesBranchpredictionetc.Simplified(reduced)instructionset14LargeRegisterFileSoftwaresolutionRequirecompilertoallocateregistersAllocatebasedonmostusedvariablesinagiventimeRequiressophisticatedprogramanalysisHardwaresolutionHavemoreregistersThusmorevariableswillbeinregisters15RegistersforLocalVariablesStorelocalscalarvariablesinregistersReducesmemoryaccessEveryprocedure(function)callchangeslocalityParametersmustbepassedResultsmustbereturnedVariablesfromcallingprogramsmustberestored16RegisterWindowsOnlyfewparametersLimitedrangeofdepthofcallUsemultiplesmallsetsofregistersCallsswitchtoadifferentsetofregistersReturnsswitchbacktoapreviouslyusedsetofregisters17RegisterWindowscont.ThreeareaswithinaregistersetParameterregistersLocalregistersTemporaryregistersTemporaryregistersfromonesetoverlapparameterregistersfromthenextThisallowsparameterpassingwithoutmovingdatacont.

1.內(nèi)容，所含之物(contents)2.繼續(xù)的；不斷的；連續(xù)的18OverlappingRegisterWindows……19CircularBufferdiagram主程序1子程序A2子程序B3子程序C4子程序D5子程序E6子程序F7子程序G20OperationofCircularBufferWhenacallismade,acurrentwindowpointerismovedtoshowthecurrentlyactiveregisterwindowIfallwindowsareinuse,aninterruptisgeneratedandtheoldestwindow(theonefurthestbackinthecallnesting)issavedtomemoryAsavedwindowpointerindicateswherethenextsavedwindowsshouldrestoreto21GlobalVariablesAllocatedbythecompilertomemoryInefficientforfrequentlyaccessedvariablesHaveasetofregistersforglobalvariables22RegistersvCacheLargeRegisterFileCacheAlllocalscalarsRecently-usedlocalscalarsIndividualvariablesBlocksofmemoryCompiler-assignedglobalvariablesRecently-usedglobalvariablesSave/RestorebasedonprocedurenestingdepthSave/RestorebasedoncachereplacementalgorithmRegisteraddressingMemoryaddressing23ReferencingaScalar-

WindowBasedRegisterFile24ReferencingaScalar-Cache25ReferencingaScalar-

WindowBasedRegisterFile26CompilerBasedRegisterOptimizationAssumesmallnumberofregisters(16-32)OptimizinguseisuptocompilerHLLprogramshavenoexplicitreferencestoregistersusually-thinkaboutC-registerintAssignsymbolicorvirtualregistertoeachcandidatevariableMap(unlimited)symbolicregisterstorealregistersSymbolicregistersthatdonotoverlapcansharerealregistersIfyourunoutofrealregisterssomevariablesusememory27GraphColoringGivenagraphofnodesandedgesAssignacolortoeachnodeAdjacentnodeshavedifferentcolorsUseminimumnumberofcolorsNodesaresymbolicregistersTworegistersthatareliveinthesameprogramfragmentarejoinedbyanedgeTrytocolorthegraphwithn

colors,wherenisthenumberofrealregistersNodesthatcannotbecoloredareplacedinmemory28GraphColoringApproach29GraphColoringApproach30WhyCISC(1)?Compilersimplification?

Disputed…ComplexmachineinstructionshardertoexploitOptimizationmoredifficult

(開發(fā))Smallerprograms?Programtakesuplessmemorybut…MemoryisnowcheapMaynotoccupylessbits,justlookshorterinsymbolicformMoreinstructionsrequirelongerop-codesRegisterreferencesrequirefewerbits

dispute

英音：[di‘spju:t]爭論;爭執(zhí)

simplification

1.單純化2.簡單化31WhyCISC(2)?Fasterprograms?BiastowardsuseofsimplerinstructionsMorecomplexcontrolunitMicroprogramcontrolstorelargerthussimpleinstructionstakelongertoexecuteItisfarfromclearthatCISCistheappropriatesolution

Bias傾向,趨勢32RISCCharacteristicsOneinstructionpercycleRegistertoregisteroperationsFew,simpleaddressingmodesFew,simpleinstructionformatsHardwireddesign(nomicrocode)FixedinstructionformatMorecompiletime/effort33RISCvCISCNotclearcutManydesignsborrowfrombothphilosophiese.g.PowerPCandPentiumIIphilosophies

哲學；觀點34RISCPipeliningMostinstructionsareregistertoregisterTwophasesofexecutionI:InstructionfetchE:ExecuteALUoperationwithregisterinputandoutputForloadandstoreI:InstructionfetchE:ExecuteCalculatememoryaddressD:MemoryRegistertomemoryormemorytoregisteroperation35Figure13.6cFigure13.6c,threeinstructionscanbeoverlapped,andtheimprovementisasmuchasafactorof3.I36Figure13.6dE1:RegisterfilereadE2:ALUoperationandregisterwriteI37EffectsofPipelining38OptimizationofPipeliningDelayedbranchDelayedLoadLoopUnrollingunroll

展開，打開（卷著的東西）39OptimizationofPipelining

（1）LoopUnrollingReplicatebodyofloopanumberoftimesIterateloopfewertimesReducesloopoverheadIncreasesinstructionparallelismImprovedregister,datacacheorTLBlocality

unroll

展開，打開（卷著的東西）

replicate

英音：[‘replikeit]折疊;復制

iterate

英音：['it?reit]反復,重復

overhead英音：[‘?uv?’hed]日常開支，額外開銷40OptimizationofPipelining

（2）DelayedbranchDoesnottakeeffectuntilafterexecutionoffollowinginstructionThisfollowinginstructionisthedelayslot41NormalandDelayedBranchAddressNormalBranchDelayedBranchOptimizedDelayedBranch100LOAD X,rALOAD X,rALOADX,rA101ADD 1,rAADD 1,rAJUMP 105102JUMP 105JUMP 106ADD 1,rA103ADD rA,rBNOOP

ADD rA,rB104SUB rC,rBADD rA,rBSUB rC,rB105STORErA,Z

SUB rC,rBSTORErA,Z

106

STORErA,Z

42UseofDelayed

BranchAddressNormalBranchDelayedBranchOptimizedDelayedBranch100LOAD X,rALOADX,rALOADX,rA101ADD1,rAADD1,rAJUMP105102JUMP105JUMP106ADD1,rA103ADDrA,rBNOOP ADDrA,rB104SUBrC,rBADDrA,rBSUBrC,rB105STORErA,Z