ACA13-lecture02-ISA

AdvancedComputerArchitecture(ACA2013)Dr.Chun-yuanZHANGCollegeofComputer,NUDTFall,2013Lecture2ISA&BasicCPUOctober11,2013CollegeofComputer,NUDTACA13,Fall3Mottos•Chapter1:BillGates,February24,2004–Ithinkit’sfairtosaythatpersonalcomputershavebecomethemostempoweringtoolwe’veevercreated.They’retoolsofcommunication,they’retoolsofcreativity,andtheycanbeshapedbytheiruser.•AppendixA:SelectionfromtheListof18MachineInstructionsfortheEDSAC(1949)–AnAddthenumberinstoragelocationnintotheaccumulator–EnIfthenumberintheaccumulatorisgreaterthanorequaltozeroexecutenexttheorderwhichstandsinstoragelocationn;otherwiseproceedserially.–ZStopthemachineandringthewarningbell.CollegeofComputer,NUDTACA13,Fall4DeeperReading:BooksofMIPSCPU•DavidMoneyHarrisandSarahL.Harris:DigitalDesignandComputerArchitecture,2007byElsevierInc.–数字设计和计算机体系结构（英文版），-北京：机械工业出版社，2007.10–HowtobuildaMIPSprocessorofH&PwithHDL(VHDLandVerilog)inthesecondhalfofthetextbook•DominicSweetman:SeeMIPSRun...Linux,secondedition,2007byElsevierInc.–MIPS体系结构透视（英文版·第2版），-北京：机械工业出版社，2007.2–UnderstandingMIPS(microprocessor)fromapeogrammer'spointofview(oldarchitecture)CollegeofComputer,NUDTACA13,Fall5BooksLookLikeCollegeofComputer,NUDTACA13,Fall6CollegeofComputer,NUDTACA13,Fall7SummarizingPerformance•Whichsystemisfaster?SystemRate(Task1)Rate(Task2)A1020B2010CollegeofComputer,NUDTACA13,Fall8…DependsWho’sSellingSystemRate(Task1)Rate(Task2)A1020B2010Average1515AveragethroughputSystemRate(Task1)Rate(Task2)A0.502.00B1.001.00Average1.251.00ThroughputrelativetoBSystemRate(Task1)Rate(Task2)A1.001.00B2.000.50Average1.001.25ThroughputrelativetoACollegeofComputer,NUDTACA13,Fall9ArchitectureandMicro-InstructionSetPrinciplesWehavereviewedChapter1InstructionSetArchitecture(ISA)—theportionofthecomputervisibletotheprogrammerorcompilerwriterCollegeofComputer,NUDTACA13,Fall11AppendixAInstructionSetPrinciples•A.1Introduction•A.2ClassifyingInstructionSetArchitectures•A.3MemoryAddressing•A.4TypeandSizeofOperands•A.5OperationsintheInstructionSet•A.6InstructionsforControlFlow•A.7EncodinganInstructionSet•A.8CrosscuttingIssues:TheRoleofCompilers•A.9PIAT:TheMIPSArchitecture•A.10FallaciesandPitfalls•A.11ConcludingRemarks•A.12HistoricalPerspectiveandReferencesCollegeofComputer,NUDTACA13,Fall12FocusonFourTopicsinTheSection•Taxonomyofinstructionsetalternatives–Givesomequalitativeassessmentoftheadvantagesanddisadvantagesofvariousapproaches•Presentandanalyzesomeinstructionsetmeasurementsthatarelargelyindependentofaspecificinstructionset•Theissueoflanguagesandcompilersandtheirbearingoninstructionsetarchitecture•TheMIPSinstructionset–AppendixK:11examplesofISACollegeofComputer,NUDTACA13,Fall13ApplicationAreasandRequirementsofISA•Manydifferentmicroarchitectureexitsforasinglearchitecture–Optimizedortrade-offs•TheISAverysimilar,butdifferent–Desktopcomputing•EmphasizesintegerandFP•Littleregardforprogramsizeorchippower–Server•FPislessimportantthanintandcharstring–PMDandEmbeddedapplication•Cost,powerandcodesizeCollegeofComputer,NUDTACA13,Fall14TraditionalTaxonomyofInst.Sets:AC•AccumulatorarchitectureALUMemoryACollegeofComputer,NUDTACA13,Fall15TraditionalTaxonomyofInst.Sets:Stack•Stackarchitecture:stackProcessorMainStoreALUMemorytopCollegeofComputer,NUDTACA13,Fall16AStackMachine•AStackmachinehasastackasapartoftheprocessorstate•typicaloperations:–push,pop,+,*,...•specifythetop2elementsofthestackasoperandsabapushbcbapushcbapopCollegeofComputer,NUDTACA13,Fall17EvaluationofExpressions(a+b*c)/(a+d*c-e)/+*+ae-acdc*bCollegeofComputer,NUDTACA13,Fall18EvaluationofExpressions:ReversePolish(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/CollegeofComputer,NUDTACA13,Fall19EvaluationofExpressions:pusha(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/pushaEvaluationStackCollegeofComputer,NUDTACA13,Fall20aEvaluationofExpressions:ainstack(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/EvaluationStackCollegeofComputer,NUDTACA13,Fall21aEvaluationofExpressions:pushb(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/pushbEvaluationStackCollegeofComputer,NUDTACA13,Fall22abEvaluationofExpressions:pushc(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/pushcEvaluationStackCollegeofComputer,NUDTACA13,Fall23abcEvaluationofExpressions:multiply(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/multiplyEvaluationStackCollegeofComputer,NUDTACA13,Fall24abcEvaluationofExpressions:operation(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/multiply*EvaluationStackCollegeofComputer,NUDTACA13,Fall25abcEvaluationofExpressions:resultonstack(a+b*c)/(a+d*c-e)/+*+ae-acdc*bReversePolishabc*+adc*+e-/*EvaluationStackb*cCollegeofComputer,NUDTACA13,Fall26TraditionalTaxonomyofInst.Sets:Registers•Registerarchitecture–Register-memory(CISC)–Register-register(RISC,load-store)–Memory-memory(notusedanymore)ALUMemoryregsCollegeofComputer,NUDTACA13,Fall27HowManyRegistersAresufficient?•First,registers—likeotherformsofstorageinternaltotheprocessor—arefasterthanmemory•Second,registersaremoreefficientforacompilertousethanotherformsofi