building a cpu - swarthmore collegebryce/cs31/f16/slides/w03b_cpu.pdf · data memory we’re...

BuildingaCPU9/15/16

opcode

X

Y

XopY

flags

AbstractionofyourLab3ALU:

InsideyourLab3ALU:

opcode

X

Y

XopY

flags

+

-

<<

…MUX

FlagLogic

CircuitsinsidetheALU

• Arithmeticcircuits• Generallyonecircuitforeachpossibleoperation.

• Controlcircuits• Selecttherightoutput• Setappropriateflags

CircuitsaroundtheALU

• Wheredotheinputscomefrom?• X,Y• opcode

• Wheredotheoutputsgo?• XopY• Flags

Recallfromlasttime…ThreemainclassificationsofHWcircuits:1. ALU:implementarithmetic&logicfunctionality

(ex)addertoaddtwovaluestogether

2. Storage:tostorebinaryvalues(ex)RegisterFile:setofCPUregisters

3. Control:support/coordinateinstructionexecution(ex)fetchthenextinstructiontoexecute

CircuitsarebuiltfromLogicGateswhicharebuiltfromtransistors

HW CircuitsLogic GatesTransistor

Recallfromlasttime…ThreemainclassificationsofHWcircuits:

2. Storage:tostorebinaryvalues(ex)RegisterFile:setofCPUregisters

HW CircuitsLogic GatesTransistor

GivetheCPUa“scratchspace”toperformcalculationsandkeeptrackofthestateitsin.

MemoryCircuits:StartingSmall

• Storea0or1

• Retrievethe0or1valueondemand(read)

• Setthe0or1valueondemand(write)

R-SLatch:StoresValueQWhenRanSareboth1:Storeavalue

RandSareneverbothsimultaneously0

• Towriteanewvalue:• SetSto0momentarily(Rstaysat1):towritea1• SetRto0momentarily(Sstaysat1):towritea0

Q (valuestored)

~Q

S

R

R-SLatch

a

b

GatedDLatchControlsS-Rlatchwriting,ensuresS&Rneverboth0

D:datawewanttostoreWE:write-enable:allowdatatobestored

Latchesusedinregisters(upnext)andSRAM(caches,later)Fast,notverydense,expensive

DRAM:capacitor-based:

Q (valuestored)

~Q

S

R

R-SLatchD

WE

WhatgetsstoredwhenWE=1?

A. Q=0B. Q=1C. Q=DD. Q=~DE. Somethingelse.

Q

~Q

S

R

D

WE

Registers

• Fixed-sizestorage(8-bit,32-bit,etc.)

• GatedDlatchletsusstoreonebit• ConnectNofthemtothesamewrite-enablewire!

Write-enable:

N-bitinputwires(bus):

N-bitRegisterBit0

Bit1

BitN-1

…

“Registerfile”• AsetofregistersfortheCPUtostoretemporaryvalues.

• You(theprogrammer)candirectlyinteractwiththeregisterfile.

• Instructionsofform:• “addR1+R2,storeresultinR3”

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

MemoryCircuitSummary

• Lotsofabstractiongoingonhere!• Gateshidethedetailsoftransistors.• BuildR-SLatchesoutofgatestostoreonebit.• CombiningmultiplelatchesgivesusN-bitregister.• GroupingN-bitregistersgivesusregisterfile.

• Registerfile’ssimpleinterface:• ReadRx’svalue,useforcalculation• WriteRy’svaluetostoreresult

Recallagain…ThreemainclassificationsofHWcircuits:1. ALU:implementarithmetic&logicfunctionality

(ex)addertoaddtwovaluestogether

2. Storage:tostorebinaryvalues(ex)RegisterFile:setofCPUregisters

3. Control:support/coordinateinstructionexecution(ex)fetchthenextinstructiontoexecute

CircuitsarebuiltfromLogicGateswhicharebuiltfromtransistors

HW CircuitsLogic GatesTransistor

Recallagain…ThreemainclassificationsofHWcircuits:

3. Control:support/coordinateinstructionexecution(ex)fetchthenextinstructiontoexecute

HW CircuitsLogic GatesTransistor

Keeptrackofwhereweareintheprogram.Executeinstruction,movetonext.

CPUsofar…

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Weknowhowtostoredata(inregisterfile).Weknowhowtoperformarithmeticonit,byfeedingittoALU.Remainingquestions:

Whichregister(s)doweuseasinputtoALU?WhichoperationshouldtheALUperform?Towhichregistershouldwestoretheresult?

Allthisinfocomesfromourcompiledprogram:aseriesofinstructions.

Recall:VonNeumannModel

CPU(ControlandArithmetic)

Input/Output

ProgramandData

Memory

We’rebuildingthis.Ourprogram(instructions)livehere.We’llassumefornowthatwecanaccessitlikeanarray.

0:

1:

2:

3:

4:

…

N-1:

Mem Addresses(buckets)

CPUGamePlan

• Fetchinstructionfrommemory

• Decodewhattheinstructionistellingustodo• TelltheALUwhatitshouldbedoing• Findthecorrectoperands

• Executetheinstruction(arithmetic,etc.)

• Storetheresult

ProgramState

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Let’saddtwomorespecialregisters(notinregisterfile)tokeeptrackofprogram.

ProgramCounter(PC): Memoryaddressofnextinstr 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): Instruction contents(bits)

Fetchinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

LoadIRwiththecontentsofmemoryattheaddressstoredinthePC.

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): InstructionatAddress0

Decodinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

Decodinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

OPCodetellsALUwhichoperationtoperform.

Decodinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

RegisterID#’sspecifyinputarguments.

Executinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

LettheALUdoitsthing.(e.g.,Add)

Storingresults.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

We’vejustcomputedsomething.Wheredoweputit?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

ResultlocationspecifieswheretostoreALUoutput.

Whydoweneedaprogramcounter?Can’twejuststartat0andcountuponeatatimefromthere?

A. Wedon’t,it’sthereforconvenience.

B. SomeinstructionsmightskipthePCforwardbymore

thanone.

C. SomeinstructionsmightadjustthePCbackwards.

D. WeneedthePCforsomeotherreason(s).

Storingresults.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

Resultmightbe:MemoryRegisterPC

RecapCPUModel

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Fourstages:fetchinstruction,decodeinstruction,execute,storeresult

ProgramCounter(PC): Memoryaddressofnextinstr 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): Instruction contents(bits)

Fetchinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

LoadIRwiththecontentsofmemoryattheaddressstoredinthePC.

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): InstructionatAddress0

Decodinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

Decodinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

OPCodetellsALUwhichoperationtoperform.

Decodinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

RegisterID#’sspecifyinputarguments.

Executinginstructions.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Whatoperation?Whicharguments?

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

LettheALUdoitsthing.(e.g.,Add)

Storingresults.

32-bitRegister#0WEDatain

32-bitRegister#1WEDatain

32-bitRegister#2WEDatain

32-bitRegister#3WEDatain

…

MUX

MUX

RegisterFile

ALU

Interprettheinstructionbits:Storeresultinregister,memory,PC.

ProgramCounter(PC): Address0 0:

1:

2:

3:

4:

…

N-1:

(Memory)

InstructionRegister(IR): OPCode|Reg A|Reg B|Result

Resultmightbe:MemoryRegisterPC

Clocking

• Needtoperiodicallytransitionfromoneinstructiontothenext.

• Ittakestimetofetchfrommemory,forsignaltopropagatethroughwires,etc.• Toofast:don’tfullycomputeresult• Tooslow:wastetime

ClockDrivenSystem• Everythinginisdrivenbyadiscreteclock• clock:anoscillatorcircuit,generateshilowpulse• clockcycle:onehi-lowpair

• Clockdetermineshowfastsystemruns• Processorcanonlydoonethingperclockcycle

• Usuallyjustonepartofexecutinganinstruction• 1GHzprocessor:

1billioncycles/secondà 1cycleeverynanosecond

Clock

1cycle

1 0 1 0 1 0 1 0 1 0

ClockandCircuitsClockEdgesTriggersevents• Circuitshavecontinuousvalues• RisingEdge:triggernewinputvalues• FallingEdge:consistentoutputreadytoread• Betweenrisingandfallingedgecanhaveinconsistentstateasnewinputvaluesflowthroughcircuit

^ newinput

^ outputready

^ newinput

Clock:

Timeperinstruction:LaundryAnalogy

• Discretestages:fetch,decode,execute,store

• Analogy(laundry):washer,dryer,folding,dresser

W Dy F Dr

4 Hours

LaundryW Dy F Dr

4 Hours

W Dy F Dr

4 Hours

W Dy F Dr

4 Hours

4-hourcycletime.

Finishesalaundryloadeverycycle.

(6laundryloadsperday)

Pipelining(Laundry)

DyW

FDyW

DrFDyW

DrFDyW

W

1Hour

1st hour:

2nd hour:

3rd hour:

4th hour:

5th hour:

Steadystate:Oneloadfinisheseveryhour!(Noteveryfourhourslikebefore.)

DF

EDF

SEDF

SEDF

F

1Nanosecond

1st nanosecond:

2nd nanosecond:

3rd nanosecond:

4th nanosecond:

5th nanosecond:

Steadystate:Oneinstructionfinisheseverynanosecond!(Clockratecanbefaster.)

CPUStages:fetch,decode,execute,storeresults

Pipelining(CPU)

Pipelining

(Formoredetailsaboutthisandtheotherthingswetalkedabouthere,takearchitecture.)

Comingupnextweek…

• TalkingtotheCPU:Assemblylanguage