国嵌2参考芯片chxorgmips微处理器研究开发中心

上讲总结:使用通用寄存器的load-store cement(withanaddressoffsetsizeof1216bits)immediatesize8to16bitsregistersubtract、moveregister-r pareequal、comparenotequal、branch(withaPC-relativeaddressatleast8-bitslong)、jump、call,以及return;支持如下数据大小和类型：8位、16位、323264IEEE754 固定指令编码方计算机科学技术 微处理器研究开发中

°更课 ,SiliconGraphics,计算机科学技术 微处理器研究开发中 MIPS

°DesignPrinciple:simplicityfavors°OfcoursethiscomplicatessomeC A=B+C+E=F-

MIPS代码

A=B+add$s0,$s1,(编译器完成寄存器与变量的关联

MIPScode:add$t0,$s1,add$s0,$t0,sub$s4,$s5,°Operandsmustberegisters,only32registers°DesignPrinciple:smalleris °Arithmeticinstructionsoperandsmustbeonly32registers°Compilerassociatesvariableswith°Whataboutprogramswithlotsof

Memory°Viewedasalarge,single-dimensionarray,withan°Amemoryaddressisanindexintothe°"Byteaddressing"meansthattheindexpointsabyteof88bitsof8bitsof8bitsof8bitsof8bitsof8bitsof8bitsof12345计算机科

微处理器研究开发

计算机科

微处理器研究开发Memory°Bytesarenice,butmostdataitemsuselarger°ForMIPS,awordis32bitsor43232bitsof32bitsof32bitsof32bitsof

°LoadandstoreC A[8]=h+48

Registershold32bitsof

MIPScode:lw$t0,add$t0,$s2,sw$t0,°232byteswithbyteaddressesfrom0to232-°230wordswithbyteaddresses0,4,8,...232-°Wordsarei.e.,whataretheleast2significantbitsofaword

°Storewordhasdestination°Rememberarithmeticoperandsareregisters,notOurFirst°Canwefigureoutthe

Sofarwehaveswap(intv[],int{inttemp=v[k]v[k]=v[k+1];v[k+1]=}

loadingwordsbutaddressingarithmeticonregisters muli$2,$5,4add$2,$4,$2lw$15,0($2)lw$16,4($2)sw$16,0($2)sw$15,4($2)jr$31

add$s1,$s2,$s3sub$s1,$s2,$s3lw$s1,100($s2)sw$s1,100($s2)

$s1=$s2+$s1=$s2-$s1=Memory[$s2+100]=计算机科学技术 微处理器研究开发中 计算机科学技术 微处理器研究开发中Machine Machine°Instructions,likeregistersandwordsofdata,arealso32bitsExample:add$t0,$s1,registershavenumbers,$t0=9,$s1=17,°Instruction°Canyouguesswhatthefieldnamesstand

°Considertheload-wordandstore-wordWhatwouldtheregularityprinciplehaveusNewprinciple:Gooddesigndemandsa°IntroduceanewtypeofinstructionI-typefordatatransferotherformatwasR-typefor°Example:lw$t0,916bitWhere'sthe °Instructionsare°Programsarestoredinmemorytobereadorwrittenjustlikememoryfordata,

°Decisionmakingalterthecontroli.e.,changethe"next"instructiontobe°Fetch&Execute

compilers,editors,

°MIPSconditionalbranchbne$t0,$t1,beq$t0,$t1,InstructionsarefetchedandputintoaspecialBitsintheregister"control"thesubsequentFetchthenextinstructionand计算机科学技术 微处理器研究开发中

if(i==j)h=i+bne$s0,$s1,add$s3,$s0,$s1 计算机科

微处理器研究开发°MIPSunconditionalbranchj

Soadd$s1,$s2,$s3sub$s1,$s2,$s3lw$s1,100($s2)sw$s1,100($s2)bne$s4,$s5,L

$s1=$s2+$s1=$s2-$s1=Memory[$s2+100]=Nextinstr.isatLabelif$s4if

beq$s4,$s5,Lab1add$s3,$s4,$s5jLab2Lab1:sub$s3,$s4,$s5

beqj

Nextinstr.isatLabelif$s4=$s5Nextinstr.isatLabel16bit26bitIJ°CanyoubuildasampleforIJControl°Wehave:beq,bne,whataboutBranch-if-less-°Newslt:setonless°slt$t0,$s1,$s2if$s1<$s2then$t0=$t0=°Canusethisinstructiontobuild"blt$s1,$s2,cannowbuildgeneralcontrol°Notethattheassemblerneedsaregistertodothis,thereare ofuseconventionsforregisters

ofUse0计算机科

微处理器研究开发中

计算机科学技术 微处理器研究开发中

Howaboutlarger°Smallconstantsareusedquitefrequently(50%of A=A+B=B+C=C-

°We'dliketobeabletoloada32bitconstantintoa°Mustusetwoinstructions,new"loadupperimmediate"lui$t0,°Solutions?Whyput'typicalconstants'inmemoryandloadcreatehard-wiredregisters(like$zero)forconstantslike°MIPSaddi$29,$29,4slti$8,$18,10andi$29,$29,6ori$29,$29,°Howdowemakethis

°Thenmustgetthelowerorderbitsright,ori$t0,

filledwith

AddressesinBranchesand°Assemblyprovidesconvenientsymbolicmucheasierthanwritingdowne.g.,destination°Machinelanguageistheunderlyinge.g.,destinationisnolonger°Assemblycanprovide'pseudoe.g.,Move$t0,$t1existsonlyinwouldbeimplementedusingAdd°Whenconsideringperformanceyoushouldcountrealinstructions

bne$t4,$t5,LabelNextinstructionisatLabelif$t4$t5beq$t4,$t5,LabelNextinstructionisatLabelif$t4=$t5jLabel NextinstructionisatLabel16bit26bit°Addressesarenot32-Howdowehandlethiswithloadandstore计算机科学技术 微处理器研究开发中 计算机科学技术 微处理器研究开发中bne$t4,$t5,LabelNextinstructionisatLabelif$t4

MIPSR2000/R3000的寄00beq$t4,$t5,LabelNextinstructionisatLabelif$t416bit°Couldspecifyaregister(likelwandsw)andaddittouseInstructionAddressRegister(PC=programmostbranchesarelocal(principleof°Jumpinstructionsjustusehighorderbitsofaddressboundariesof256

°可编232x31x32-bitGPRs(R032x32-bitFPHI,LO,

..MIPS的寻址方式/指令Register++

MIPSR2000/R3000的操Add,AddU, SubU,And,Or,Nor,SLT,AddI,AddIU,SLTI,SLTIU,AndI,OrI,XorI,SLL,SRL,SRA,SLLV,SRLV, (MemoryLB,LBU,LH,LHU,LW,SB,SH,SW,SWL,计算机科学技术 微处理器研究开发中 计算机科学技术 微处理器研究开发中°Startmultiply,MULTrs,MULTUrs,DIVrs,DIVUrs,°MovefromHIorMFHIMFLO°MovetoHIorMTHIMTLO

MIPS的算术 add$1,$2,$3 $1=$2+$3 3operands;exceptionpossible sub$1,$2,$3 $1=$2-$3 3operands;exceptionpossibleaddimmediate addi$1,$2,100 $1=$2+100 +constant;exceptionpossibleaddunsigned addu$1,$2,$3 $1=$2+$3 3operands;noexceptionssubtractunsigned subu$1,$2,$3 $1=$2-$3 3operands;noexceptionsaddimm.unsign. addiu$1,$2,100 $1=$2+100 +constant;noexceptions mult$2,$3 Hi,Lo=$2x$3 64-bitsignedproductmultiplyunsigned Hi,Lo=$2x$3 64-bitunsignedproduct div$2,$3 Lo=$2÷$3, Lo=quotient,Hi=remainderHi=$2moddivide divu Lo=$2 Unsignedquotient&Hi=$2modMovefromHi mfhi$1 $1=Hi UsedtogetcopyofHiMovefrom mflo $1= UsedtogetcopyofMIPS的逻辑

OperandSizeandand$1=$2&3reg.operands;Logicalor$1=$2|3reg.operands;Logicalxor$1=$23reg.operands;Logicalnor$1=~($23reg.operands;Logicalandandi$1=$2&LogicalANDreg,orori$1=$2|LogicalORreg,xorxori$1,$1=~$2LogicalXORreg,shiftleftsll$1=$2<<Shiftleftbyshiftrightsrl$1=$2>>Shiftrightbyshiftrightsra$1=$2>>Shiftright(signshiftleftsllv$1=$2<<Shiftleftbyshiftrightsrlv$1,$2,$1=$2>>Shiftrightbyshiftrightsrav$1,$2,$1=$2>>Shiftrightarith.by

IntIntFP Frequencyofreferenceby

计算机科学技术 微处理器研究开发中

Supportthesedatasizesand8-bit,16-bit,32-bitintegers32-bitand64-bitIEEE754floatingpoint计算机科学技术 微处理器研究开发中MIPS的数据

ConditionalBranchSW500(R4),R3SH502(R2),R3SB41(R3),R2LWR1,

StorewordStorehalfStorebyteLoad

PC-relativesincemostbranchestothecurrentPCaddressAtleast8bitssuggested(128IntFPCompareEqual/NotEqualmostimportantIntFPLHR1,40(R3)LHUR1,40(R3)LBR1,40(R3)LBUR1,40(R3)

LoadLoadhalfwordLoadLoadbyte

LUIR1, LoadUpperImmediate(16bitsshiftedleftby计算机科学技术 微处理器研究开发中

Frequencyof计算机科学技术系typesin

微处理器研究开发°比较并转

BEQrs,rt,BNErs,rt,°与零比较并转BLEZrs,BGTZrs,BLTZALrs,

ifR[rs]==R[rt]thenPC-relativeifR[rs]<=0thenPC-relative><IfR[rs]<0thenbranchandlink(intoR

branchonequalbranchonnoteq.setonlessthansetlessthanimm.setlessthanuns.setl.t.imm.uns.

beqEqualtest;PCrelativebneNotequaltest;PCslt;slti<;sltu;sltiu;

if($1==$2)gotoif($1!=$2)goto if($2<$3)$1=1;else$1=0if($2<100)$1=1;else$1=0if($2<$3)$1=1;else$1=0if($2<100)$1=1;else$1=0°其他的比较并转移需要两条指

jjump jr

gotogoto°几乎所有的比较是计算机科

进行比

微处理器研究开发

jumpand jal $31=PC+4;gotoprocedure计算机科学技术 微处理器研究开发中

从程 系统中获得指

确定后续指

branchonequalbranchonnoteq.

beqEqualtest;PCrelativebneNotequaltest;PC

if($1==$2)gotoif($1!=$2)goto 确定所需的动作和指令

定位并获得操作数数计算结果数值或状 系统中存放结果，以备后

branchon beqbranchonnot bne

if($1==$2)goto Equaltest;PCrelativeif($1!=$2)goto Notequaltest;PCAACBABACBABA

器堆栈(Memory即使操作数堆栈不是系统结构的组成部分,也有益于栈式环境(stacked以 子程序调用和返 CALL

堆栈为空表明什么

inf.abcabc0

0 inf.一些机器提供一 器堆栈(memorystack),并将此作为系统结构的一部

BAALittle-->Big/LastBAA

Little-->Big/Next(例如,一些机器堆栈是利用软件来实(例如,

ReadfromMem(SP)DecrementSPIncrementSPWritetoMem(SP)

ReadfromWritetoIncrement计算机科学技术 微处理器研究开发中

计算机科

微处理器研究开发 :Stack

High

(oldFP,Local(oldFP,Localfrom2expressionevaluation3function456

zeroconstant for

16s0callee...(callercan在表达式估值过程中,伸

temporary(caller77Pointertoglobal8Stack°堆栈有许多可能的变种(up/down,lastpushed/next

Low

.. (calleecan15

30fpframe31raReturnAddress°BlockstructuredlanguagescontainlinktolexicallyenclosingMIPS/addiu$sp,$sp,-32 addiu$fp,$sp,.. $a0,

C示例:swap(intv[],int{inttemp;temp=v[k];v[k]=v[k+1];v[k+1]=temp;}lwlw$fp,16($sp)addiu$sp,$sp,32jr$31计算机科

old微处理器研究开发

$16是中间使用的寄存器(scratchMIPS代码计算机科学技术

微处理器研究开发

Branch&lir3,#7subr4,r4,addir5,r3,

Delay$sp,$sp,-;create$sp,$sp,-;createspaceon$16,;calleesavedregisterputonto$t2,;mulitplykby$t2,;addressof$15,;load$16,;load$16,;storev[k+1]into$15,;storeoldvalueofv[k]into$16,;calleesavedregisterrestoredfrom$sp,$sp,;restoretopof

Branch ;return cethatcalled

BytheendofBranchinstruction,theCPUknowswhetherornotthebranchwilltake However,itwillhavefetchedthenextinstructionbyregardlessofwhetherornotabranchwillbetaken.Whynotexecuteit?FillingDelayed

&Dcd&Dcd&OpInstInstexecutesuccessorevenifbranchInstThenbranchor

Dcd&Dcd&OpInstSingledelayslotimpactsthecriticalpath

Compilercanfillasingledelayslotwithausefulinstruction50%ofthetrytomovedownfromabove

addr3,r1,r2subr4,r4,1bzr4,LL

,moveupfromtarget,if addrd,Isthisviolatingthe

无论这个转移是否真的发生，它都将访取下一条指令计算机科学技术 微处理器研究开发中 计算机科学技术 微处理器研究开发中Dcd&Dcd&OpInst

MIPS的其他指Dcd&OpInstDcd&OpInstInstThenbranchorcontinue Singledelayslotimpactsthecriticalpath

°coprocessor°TLB

Abreakpointtrapoccurs,transferscontroltoexceptionhandlerAsystemtrapoccurs,transferscontroltoexceptionhandlerSupportforfloatingpoint:discussedlaterSupportforvirtualmemory:discussedlater试图从跳转之前，移动下来指如果安全，可以从目标出移动来指

addr3,r1,subr4,r4,1bzr4,LLLL:addrd,

°restorefrom°loadword°storeword

Restorespreviousinterruptmask&kernel/usermodebitsintostatusregisterSupportsmisalignedwordloadsSupportsmisalignedwordstores°#0寄存器总是具有数值0(即使试图对它写入其他数值，也是如此°转移和跳转指令将返回地址PC+4装 寄存器（link°所有指令改变目标寄存器的所有32位信息(包括lui,lb,lh)，并 的所有32位信息(add,sub,and,or,等等）°立即数算术和逻辑指令进行如下扩展逻辑立即数（logical 零扩展到32算术立即数（arithmeticimmediates）符号扩展到32°指令lblh装入的数据进行如下扩展lbu,lhu是零扩lb 是符号扩°下面的算术和逻辑指令可能会出现溢出add,sub,以下指令不会出现溢出addusubu,addiuand,orxor,norshifts,mult,multu,div,divu

°Designprovidemorepowerfulgoalistoreducenumberofinstructionsdangerisaslowercycletimeand/orahigher°SometimesreferredtoasRISCvs.virtuallyallnewinstructionsetssince1982havebeenVAX:minimizecodesize,makeassemblylanguageinstructionsfrom1to54bytes°WewilllookatPowerPCand计算机科学技术 微处理器研究开发中 计算机科学技术 微处理器研究开发中°Indexed lwWhatdowehavetodoin°Update

°1978:The 8086isannounced(16bit°1980:The8087floatingpointcoprocessoris°1982:The80286increasesaddressspaceto24bits,°1985:The80386extendsto32bits,newaddressingupdatearegisteraspartofload(formarchingthroughlwu$t0,4($s3)Whatdowehavetodoinloadmultiple/storeaspecialcounterregisterbcdecrementcounter,ifnot0goto

°1989-1995:The80486,Pentium,PentiumProaddafewinsts(mostlydesignedforhigherperformance)°1997:MMXis°1997-2000:PentiumII,PentiumIII,PentiumThishistorytheimpactoftheGoldenhandcuffsofAnarchitecturethatisdifficultto inandimpossibleto 8086/88=>80286=>80386=>80486=>Pentium=>PentiumPro8086只能使用不多的晶体管，因而它只实现了16位微处理试图与8位8080微处理器比较兼在8086的后续微处理器中，不断增加了许多新的功许多不同的计算机工程师的成1978年宣°VAX简单的编&小代码空间高效的指令强大的寻址强大的寄存器一位出色设计师的1977年宣计算机科学技术 微处理器研究开发中

Adominantarchitecture:°SeeyourtextbookforamoredetailedInstructionsfrom1to17bytesoneoperandmustactasbothasourceandoneoperandcancomefromcomplexaddressinge.g.,Baseorscaledindexwith8or32bit °SavingthemostfrequentlyusedinstructionsarenottoodifficulttocompilersavoidtheportionsofthearchitecturethatareWhatthe80x86lacksinstyleismadeupin makingitbeautifulfromtheright计算机科学技术 微处理器研究开发中机器示例:地址与寄存

VAX

220x8bitbytesAX,BX,CX,

acc,index,count