1 COMP541 Sequencing – III (Sequencing a Computer) Montek Singh April 9, 2007.

1

COMP541COMP541

Sequencing – IIISequencing – III(Sequencing a Computer)(Sequencing a Computer)

Montek SinghMontek Singh

April 9, 2007April 9, 2007

2

Test 2Test 2 On April 17On April 17 Covers Covers

MemoriesMemories ArithmeticArithmetic DatapathsDatapaths SequencingSequencing

3

Design Reviews: Last Week of Design Reviews: Last Week of ClassesClasses Individual meetings during class time next Individual meetings during class time next

week (and maybe one more day)week (and maybe one more day) 20 minutes20 minutes Please prepare a presentationPlease prepare a presentation

Not necessarily a PPT, but don’t make up your Not necessarily a PPT, but don’t make up your description on the flydescription on the fly

4

Chapter 10-7Chapter 10-7 Simple computer architectureSimple computer architecture

Not unlike MIPS, except 16 bitsNot unlike MIPS, except 16 bits

Single-cycle hardwired controlSingle-cycle hardwired control Multicycle microprogrammed controlMulticycle microprogrammed control

5

Instruction FormatsInstruction Formats Register-type instructionsRegister-type instructions

Only 8 registers (3 bits)Only 8 registers (3 bits)

6

ImmediateImmediate

Only 3 bits for the immediate value (Op)Only 3 bits for the immediate value (Op) Mostly useful for typical increments/decrementMostly useful for typical increments/decrement

Or just as an exampleOr just as an example

7

BranchingBranching PC relative branchingPC relative branching

The 6 bits are sign extended to 16The 6 bits are sign extended to 16 Opcode might specify branch on zero, if Opcode might specify branch on zero, if

register SA is zeroregister SA is zero

8

Example InstructionsExample Instructions

9

Contrast to MicrooperationsContrast to Microoperations Although appear similar, they’re notAlthough appear similar, they’re not Computer instructions fetched using PCComputer instructions fetched using PC Branching much more generalBranching much more general Decoding of computer instructions usually Decoding of computer instructions usually

more complexmore complex

10

ResourcesResources Book implies Harvard Book implies Harvard

architecturearchitecture Separate I and DSeparate I and D They treat I memory as They treat I memory as

ROMROM AsynchronousAsynchronous

11

Single-Cycle ControlSingle-Cycle Control Datapath is same as example we used in Datapath is same as example we used in

datapath topicdatapath topic Next slide shows for reviewNext slide shows for review

First look at overall controlFirst look at overall control Then look at instruction decoderThen look at instruction decoder

12

Datapath & Control WordDatapath & Control Word

13

ArchArch

Instruction Memory treated as ROM (combinational)

3 bits sent to ALU for immediate instructions

Address offset sent for relative branch. Adder not shown

(neither is PC increment).

Branch control inputs

are status from ALU and

lines from decoder.

14

Instruction DecoderInstruction Decoder Many lines (the Many lines (the

three regs) need three regs) need no logicno logic RISC StyleRISC Style

Architecture Architecture tailored so parts of tailored so parts of inst. correspond to inst. correspond to control linescontrol lines

15

ControlControl Not much more to sayNot much more to say

Simple, partly because decoding so straightforwardSimple, partly because decoding so straightforward DrawbacksDrawbacks

Some instructions, like multi-cycle shifts, can’t be Some instructions, like multi-cycle shifts, can’t be implemented w/o complex datapathimplemented w/o complex datapath

Two memories (essentially a ROM and an async data Two memories (essentially a ROM and an async data memory)memory)Two cycles needed to use one memoryTwo cycles needed to use one memory

Biggest problem is delayBiggest problem is delay

16

Delay in Single-Cycle ControlDelay in Single-Cycle Control Worst case delay with reasonable componentsWorst case delay with reasonable components

Say, total 17nsSay, total 17ns Could only clock at about 50 MHzCould only clock at about 50 MHz

Pipelining is a solutionPipelining is a solution First let’s look at multi-cycle controlFirst let’s look at multi-cycle control

17

Multi-Cycle Hardwired ControlMulti-Cycle Hardwired Control GoalsGoals

Support more complex instructionsSupport more complex instructions Use single memoryUse single memory

Not necessarily coupled with multi-cycleNot necessarily coupled with multi-cycle

18

ArchArch

19

Instruction RegisterInstruction Register

IL – load IL – load signal for IRsignal for IR

PS, for PC PS, for PC controlcontrol Hold value Hold value

multiple multiple cycles, cycles, increment, increment, load, etc.load, etc.

20

Single MemorySingle Memory

PC addresses PC addresses memorymemory

Mux M gates Mux M gates addressaddress

MM signal to MM signal to select select program/data program/data addressaddress

Inst. stored in Inst. stored in IRIR

21

Added Temporary RegsAdded Temporary Regs

Now 16x16Now 16x16 8 not visible 8 not visible

to userto user New signals New signals

to address to address the registersthe registers

22

Sequence ControlSequence Control

23

Control WordControl Word

24

Control DesignControl Design Not hard to specify state diagramNot hard to specify state diagram

Derived from definition of ISADerived from definition of ISA

Hard to design logic manuallyHard to design logic manually If didn’t have logic synthesis, would probably If didn’t have logic synthesis, would probably

use microprogramminguse microprogramming

25

Two-Cycle InstructionsTwo-Cycle Instructions Simplest instructions have 2 cyclesSimplest instructions have 2 cycles

Fetch (instruction)Fetch (instruction) ExecuteExecute

This is minimum necessaryThis is minimum necessary They assume async memoryThey assume async memory

Don’t need extra clock cyclesDon’t need extra clock cycles

26

Basic Inst.Basic Inst.

27

BranchBranch Test and modify PCTest and modify PC

28

Next StepNext Step Make a table or write Verilog from ASM Make a table or write Verilog from ASM

diagram and instruction descriptionsdiagram and instruction descriptions Tedious, but not hardTedious, but not hard Same as you’ve done, with more detailsSame as you’ve done, with more details

State machine easy for these instructionsState machine easy for these instructions

29

Table from ISA and ASMTable from ISA and ASM

30

Load Register IndirectLoad Register Indirect Three cyclesThree cycles

Temporary register usedTemporary register used

31

ShiftShift Shift right/left multipleShift right/left multiple R[SA] to be shiftedR[SA] to be shifted First tested for 0First tested for 0 R9 loaded with shift lengthR9 loaded with shift length

32

Multi-Cycle TableMulti-Cycle Table

33

Summary – multi-cycleSummary – multi-cycle Multi-cycle computer enables more complex Multi-cycle computer enables more complex

instructionsinstructions May also be fasterMay also be faster We’ll also briefly look at We’ll also briefly look at pipelinedpipelined computers – computers –

more parallelism but more complex controlmore parallelism but more complex control

34

Limits to Clock PeriodLimits to Clock Period Conventional datapathConventional datapath 12 ns delay, so maximum is 83 12 ns delay, so maximum is 83

MHz clockMHz clock Maybe have even tighter Maybe have even tighter

constraints due to control logicconstraints due to control logic

35

PipeliningPipelining Break datapath into stagesBreak datapath into stages Add registers between stagesAdd registers between stages Like production line – book Like production line – book

uses car wash exampleuses car wash example Wash, rinse, dryWash, rinse, dry

36

Latency vs ThroughputLatency vs Throughput LatencyLatency, the amount of time it takes to , the amount of time it takes to

execute an instruction, does not improveexecute an instruction, does not improve In fact, typically increasesIn fact, typically increases

Throughput, the number of instructions Throughput, the number of instructions executed per second, increasesexecuted per second, increases By almost the number of pipeline stagesBy almost the number of pipeline stages

37

Expected PerformanceExpected Performance Longest stage is 5nsLongest stage is 5ns So clock can be 200 MHzSo clock can be 200 MHz Not 3 x 83 MHz. Why?Not 3 x 83 MHz. Why?

Latency is 15nsLatency is 15ns Also extra hardwareAlso extra hardware

38

DatapathDatapath 3 Stages3 Stages

Operand FetchOperand Fetch ExecuteExecute Write BackWrite Back

Note that WB register is the Note that WB register is the register file (same as at top)register file (same as at top)

39

Pipelined ExecutionPipelined Execution

Note pipeline fill and emptyNote pipeline fill and empty Efficiency is not 100%Efficiency is not 100% Important to not Important to not stallstall pipeline pipeline