CS 61C: Great Ideas in Computer Architecture Introduction ...cs61c/fa17/lec/05/L05 RISCV Intro... · –VAX architecture had an instruction to multiply polynomials! •RISC philosophy

CS61C:GreatIdeasinComputerArchitectureIntroductiontoAssemblyLanguageand

RISC-VInstructionSetArchitectureInstructors:

Krste Asanović &RandyH.Katzhttp://inst.eecs.Berkeley.edu/~cs61c

9/7/17 Fall2017- Lecture#5 1

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 2

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 3

LevelsofRepresentation/Interpretation

lw $t0,0($2)lw $t1,4($2)sw $t1,0($2)sw $t0,4($2)

HighLevelLanguageProgram(e.g.,C)

AssemblyLanguageProgram(e.g.,RISC-V)

MachineLanguageProgram(RISC-V)

HardwareArchitectureDescription(e.g.,blockdiagrams)

Compiler

Assembler

MachineInterpretation

temp=v[k];v[k]=v[k+1];v[k+1]=temp;

0000 1001 1100 0110 1010 1111 0101 10001010 1111 0101 1000 0000 1001 1100 0110 1100 0110 1010 1111 0101 1000 0000 1001 0101 1000 0000 1001 1100 0110 1010 1111

ArchitectureImplementation

Anythingcanberepresentedasanumber,

i.e.,dataorinstructions

LogicCircuitDescription(CircuitSchematicDiagrams)

9/7/17 4

InstructionSetArchitecture(ISA)• JobofaCPU(CentralProcessingUnit,akaCore):execute

instructions• Instructions:CPU’sprimitivesoperations

– Likeasentence:operations(verbs)appliedtooperands(objects)processedinsequence…

– Withadditionaloperationstochangethesequence• CPUsbelongto“families,”eachimplementingitsownsetof

instructions• CPU’sparticularsetofinstructionsimplementsanInstructionSet

Architecture (ISA)– Examples:ARM,Intelx86,MIPS,RISC-V,IBM/MotorolaPowerPC(old

Mac),IntelIA64,...59/7/17

AssemblyLanguage High-LevelLanguage69/7/17

AssemblyLanguage High-LevelLanguage79/7/17

InstructionSetArchitectures• Earlytrend:addmoreinstructionstonewCPUsforelaborateoperations– VAXarchitecturehadaninstructiontomultiplypolynomials!

• RISCphilosophy(Cocke IBM,PattersonUCB,HennessyStanford,1980s)– ReducedInstructionSetComputing– Keeptheinstructionsetsmallandsimple,inordertobuildfasthardware

– Letsoftwaredocomplicatedoperationsbycomposingsimplerones

89/7/17

RISC-VGreenCard(intextbook)

99/7/17http://inst.eecs.berkeley.edu/~cs61c/resources/RISCV_Green_Sheet.pdf

InspiredbytheIBM360

“GreenCard”

109/7/17

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 11

WhatisRISC-V?• FifthgenerationofRISCdesignfromUCBerkeley• Ahigh-quality,license-free,royalty-freeRISCISAspecification• Experiencingrapiduptakeinbothindustryandacademia• Bothproprietaryandopen-sourcecoreimplementations• Supportedbygrowingsharedsoftwareecosystem• Appropriateforalllevelsofcomputingsystem,from

microcontrollerstosupercomputers– 32-bit,64-bit,and128-bitvariants(we’reusing32-bitinclass,textbook

uses64-bit)• Standardmaintainedbynon-profitRISC-VFoundation

129/7/17

FoundationMembers(60+)

13

Rumble Development

Platinum:

Gold,Silver,Auditors:

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 14

AssemblyVariables:Registers• UnlikeHLLlikeCorJava,assemblydoesnothave

variablesasyouknowandlovethem– Moreprimitive,closerwhatsimplehardwarecandirectly

support• Assemblyoperandsareobjectscalledregisters

– Limitednumberofspecialplacestoholdvalues,builtdirectlyintothehardware

– Operationscanonlybeperformedonthese!• Benefit:Sinceregistersaredirectlyinhardware,theyare

veryfast(fasterthan1ns- lighttravels1footin1ns!!!)

9/7/17 15

Processor

Control

Datapath

RegistersliveinsidetheProcessor

16

PC

RegistersArithmetic&LogicUnit

(ALU)

Memory Input

Output

Bytes

Enable?Read/Write

Address

WriteData

ReadData

Processor-MemoryInterface I/O-MemoryInterfaces

Program

Data

CS61c

NumberofRISC-VRegisters• Drawback:Sinceregistersareinhardware,therearealimited

numberofthem– Solution:RISC-Vcodemustbecarefullywrittentoefficientlyuseregisters

• 32registersinRISC-V,referredtobynumberx0 – x31– Registersarealsogivensymbolicnames,describedlater– Why32?Smallerisfaster,buttoosmallisbad.Goldilocksprinciple(“This

porridgeistoohot;Thisporridgeistoocold;thisporridgeisjustright”)• EachRISC-Vregisteris32bitswide(RV32 variantofRISC-VISA)

– Groupsof32bitscalledaword inRISC-VISA– P&HCoD textbookusesthe64-bitvariantRV64(explaindifferenceslater)

• x0 isspecial,alwaysholdsvaluezero– Soreallyonly31registersabletoholdvariablevalues

9/7/17 17

C,JavaVariablesvs.Registers• InC(andmostHLLs):

– Variablesdeclaredandgivenatype• Example: int fahr, celsius;

char a, b, c, d, e;

– EachvariablecanONLYrepresentavalueofthetypeitwasdeclared(e.g.,cannotmixandmatchint andchar variables)

• InAssemblyLanguage:– Registershavenotype;– Operation determineshowregistercontentsareinterpreted

9/7/17 18

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 19

RISC-VInstructionAssemblySyntax• Instructionshaveanopcode andoperands•E.g.,add x1, x2, x3 # x1 = x2 + x3

9/7/17 20

Operationcode(opcode)Destinationregister Secondoperandregister

Firstoperandregister

#isassemblycommentsyntax

AdditionandSubtractionofIntegers• AdditioninAssembly

– Example: add x1,x2,x3 (inRISC-V)– Equivalentto: a=b+c (inC)whereCvariables⇔ RISC-Vregistersare:

a⇔ x1,b⇔ x2,c⇔ x3• SubtractioninAssembly

– Example: sub x3,x4,x5 (inRISC-V)– Equivalentto: d=e- f (inC)whereCvariables⇔ RISC-Vregistersare:

d⇔ x3,e⇔ x4,f⇔ x5

9/7/17 21

AdditionandSubtractionofIntegersExample1

• HowtodothefollowingCstatement?a=b+c+d- e;

• Breakintomultipleinstructionsadd x10, x1, x2 # a_temp = b + cadd x10, x10, x3 # a_temp = a_temp + dsub x10, x10, x4 # a = a_temp - e

• AsinglelineofCmayturnintoseveralRISC-Vinstructions

9/7/17 22

Immediates• Immediates arenumericalconstants• Theyappearoftenincode,sotherearespecialinstructions

forthem• AddImmediate:

addi x3,x4,-10 (inRISC-V)f=g- 10 (inC)

whereRISC-Vregistersx3,x4 areassociatedwithCvariablesf,g• Syntaxsimilartoadd instruction,exceptthatlastargumentis

anumberinsteadofaregisteradd x3,x4,x0 (inRISC-V)f=g (inC)

9/7/17 23

Processor

Control

Datapath

DataTransfer:LoadfromandStoreto memory

PC

RegistersArithmetic&LogicUnit

(ALU)

Memory Input

Output

Bytes

Enable?Read/Write

AddressWriteData=StoretomemoryReadData=Loadfrommemory

Processor-MemoryInterface I/O-MemoryInterfaces

Program

Data

9/7/17 24

MuchlargerplaceToholdvalues,but

slowerthanregisters!

FastbutlimitedplaceToholdvalues

0123…

MemoryAddressesareinBytes• Datatypicallysmallerthan32bits,butrarelysmallerthan8bits

(e.g.,chartype)–worksfineifeverythingisamultipleof8bits• 8bitchunkiscalledabyte

(1word=4bytes)• Memoryaddressesarereally

inbytes,notwords• Wordaddressesare4bytes

apart– Wordaddressissameasaddressof

rightmostbyte– least-significantbyte(i.e.Little-endianconvention)

9/7/1725

Least-significantbyteinaword

04812…

15913…

261014…

371115…

3124 2316 158 70Least-significantbytegetsthesmallestaddress

Transferfrom MemorytoRegister• Ccode

int A[100];g = h + A[3];

• UsingLoadWord(lw)inRISC-V:lw x10,12(x13) #Reg x10getsA[3]add x11,x12,x10 #g=h+A[3]

Note: x13 – baseregister(pointertoA[0])12 – offsetinbytes

Offsetmustbeaconstantknownatassemblytime

9/7/17 26

TransferfromRegisterto Memory• Ccode

int A[100];A[10] = h + A[3];

• UsingStoreWord(sw)inRISC-V:lw x10,12(x13) #Tempreg x10getsA[3]add x10,x12,x10 #Tempreg x10getsh+A[3]sw x10,40(x13) #A[10]=h+A[3]

Note: x13 – baseregister(pointer)12,40 – offsetsinbytes

x13+12andx13+40mustbemultiplesof4

9/7/17 27

Loading and Storing Bytes• Inadditiontoworddatatransfers

(lw,sw),RISC-Vhasbytedatatransfers:– loadbyte:lb– storebyte:sb

• Sameformataslw,sw• E.g.,lb x10,3(x11)

– contentsofmemorylocationwithaddress=sumof“3”+contentsofregisterx11 iscopiedtothelowbytepositionofregisterx10.

28

byteloaded

zzz zzzzx

…is copied to “sign-extend”This bit

xxxx xxxx xxxx xxxx xxxx xxxxx10:

9/7/17

Yourturn

29

Answer x12

RED 0x5

GREEN 0xf

ORANGE 0x3

0xffffffff

addi x11,x0,0x3f5sw x11,0(x5)lb x12,1(x5)

Yourturn

30

Answer x12

RED 0x5

GREEN 0xf

ORANGE 0x3

0xffffffff

addi x11,x0,0x3f5sw x11,0(x5)lb x12,1(x5)

SpeedofRegistersvs.Memory• Giventhat

– Registers:32words(128Bytes)– Memory(DRAM):Billionsofbytes(2GBto8GBonlaptop)

• andphysicsdictates…– Smallerisfaster

• HowmuchfasterareregistersthanDRAM??• About100-500timesfaster!

– intermsoflatencyofoneaccess

319/7/17

Administrivia• HW#0duetomorrownight!• HW#1willbepublishedsoon

– Two-partCprogrammingassignment

• SmallGroupTutoringsignupswillbeoutrightaftertoday’slecture

• ThreeweekstoMidterm#1!

329/7/17

Break!

9/7/17 33

RISC-VLogicalInstructions

Logical operationsC

operatorsJava

operatorsRISC-V

instructionsBit-by-bit AND & & andBit-by-bit OR | | orBit-by-bit XOR ^ ^ xorShift left logical << << sllShift right logical >> >> srl

• Usefultooperateonfieldsofbitswithinaword− e.g.,characterswithinaword(8bits)

• Operationstopack/unpackbitsintowords• Calledlogicaloperations

9/7/17 34

Logical Shifting• ShiftLeftLogical:slli x11,x12,2 #x11=x12<<2

– Storeinx11 thevaluefromx12 shifted2bitstotheleft(theyfalloffend),inserting0’s onright;<<inC

Before:00000002hex00000000000000000000000000000010twoAfter: 00000008hex00000000000000000000000000001000two

Whatarithmeticeffectdoesshiftlefthave?

• ShiftRightLogical:srli isoppositeshift;>>–Zerobitsinsertedatleftofword,rightbitsshiftedoffend

9/7/17 35

ArithmeticShifting• Shiftrightarithmetic(srai)movesn bitstotheright(inserthigh-ordersignbitintoemptybits)

• Forexample,ifregisterx10contained11111111111111111111111111100111two=-25ten

• Ifexecutesra x10,x10,4,resultis:11111111111111111111111111111110two=-2ten

• Unfortunately,thisisNOTsameasdividingby2n− Failsforoddnegativenumbers− Carithmeticsemanticsisthatdivisionshouldroundtowards0

9/7/17 36

ComputerDecisionMaking• Basedoncomputation,dosomethingdifferent• Inprogramminglanguages:if-statement

• RISC-V:if-statementinstructionisbeq register1,register2,L1

means:gotostatementlabeledL1if(valueinregister1)==(valueinregister2)….otherwise,gotonextstatement

• beq standsforbranchifequal• Otherinstruction:bne forbranchifnotequal

9/7/17 37

TypesofBranches• Branch – changeofcontrolflow

• ConditionalBranch – changecontrolflowdependingonoutcomeofcomparison– branchifequal(beq)orbranchifnot equal(bne)– Alsobranchiflessthan(blt)andbranchifgreaterthanorequal

(bge)

• UnconditionalBranch – alwaysbranch– aRISC-Vinstructionforthis:jump(j)

9/7/17 38

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 39

Exampleif Statement• Assumingtranslationsbelow,compileif block

f→x10 g→x11 h→x12i →x13 j→x14

if (i == j) bne x13,x14,Exitf = g + h; add x10,x11,x12

Exit:• Mayneedtonegatebranchcondition9/7/17 40

Exampleif-else Statement• Assumingtranslationsbelow,compile

f→x10 g→x11 h→x12i →x13 j→x14

if (i == j) bne x13,x14,Else f = g + h; add x10,x11,x12

else j Exit f = g – h; Else: sub x10,x11,x12

Exit: 9/7/17 41

Magnitude Compares in RISC-V• Untilnow,we’veonlytestedequalities(==and!=inC);

Generalprogramsneedtotest<and>aswell.• RISC-Vmagnitude-comparebranches:

“BranchonLessThan”Syntax:blt reg1,reg2, labelMeaning: if(reg1<reg2)//treatregistersassignedintegers

goto label;• “BranchonLessThanUnsigned”

Syntax:bltu reg1,reg2, labelMeaning: if(reg1<reg2) //treatregistersasunsignedintegers

goto label;

9/7/17 42

CLoopMappedtoRISC-VAssemblyint A[20];int sum = 0;for (int i=0; i<20; i++)

sum += A[i];

add x9, x8, x0 # x9=&A[0]add x10, x0, x0 # sum=0add x11, x0, x0 # i=0Loop:

lw x12, 0(x9) # x12=A[i]add x10,x10,x12 # sum+=addi x9,x9,4 # &A[i++]addi x11,x11,1 # i++addi x13,x0,20 # x13=20blt x11,x13,Loop

43

Outline• AssemblyLanguage• RISC-VArchitecture• Registersvs.Variables• RISC-VInstructions• C-to-RISC-VPatterns• AndinConclusion…

9/7/17 44

InConclusion,…• Instructionsetarchitecture(ISA)specifiesthesetof

commands(instructions)acomputercanexecute• Hardwareregistersprovideafewveryfastvariablesfor

instructionstooperateon• RISC-VISArequiressoftwaretobreakcomplexoperationsinto

astringofsimpleinstructions,butenablesfaster,simplehardware

• Assemblycodeishuman-readableversionofcomputer’snativemachinecode,convertedtobinarybyanassembler

45