computer architecture -...

Computer ArchitectureIntroduction

S. Coudert and R. Pacalet

Telecom ParisInstitut Mines-Telecom

March 2, 2021

1 / 64 March 2, 2021 Telecom Paris Institut Mines-Telecom Computer Architecture — Introduction

Outline

1 General introduction

2 General computer architectureComponentsCommunication, interconnectionIntegration

3 Multilevel architectureHardwareArchitecture, micro-architectureInstruction Set Architecture (ISA)Assembly languageSoftware execution (brief overview)

Invoking softwareOperating System (OS)Booting a computer


Acknowledgements

David Wentzlaff (Princeton University)Agner Fog (Technical University of Denmark)David Andrew Patterson (University of California, Berkeley)John Leroy Hennessy (Stanford University)James R. Larus (École Polytechnique Fédérale de Lausanne)


Course content

CompArch

Digital

Lev. 1:

Operating system

Problem−oriented languageLev. 5:

Lev. 3:

Lev. 2: Instruction set architecture

partial interpretation

interpretation or direct execution

translation: assembler

hardware

compiler or virtual machine

Lev. 4:

Microarchitecture

Lev. 0:

Assembly languageuser

System software

Application software

Environment

Hardware


The CompArch course at a glance

Short introduction to computer architecture8 to 10 lectures, 3 hours each• discussions• lecture• exercises

4 to 5 labs, 3 hours each, 30% of final gradeMaybe 2 mini-conferences (2×1.5 hours), mandatoryExam, 2 hours, 70% of final grade• Questions• Problems• With documents (any document you like)

A WEB site: https://perso.telecom-paris.fr/pacalet/CompArch/A GitLab project for the labs: https://gitlab.eurecom.fr/renaud.pacalet/comparch


General computer architecture

Hundreds of kinds of computersWork stations, servers, desktop and laptop PCs…Multiprocessor, networked computers, clustersEmbedded systems:• Smart phones, tablets…• Application specific architectures

— Satellites— Smart cards— Automotive…

• Micro-controllers• Systems on Chip (SoC)• IoT

Many different (standard) components, combined for specific applications


Components (1)

Computing

CPU, GPU, dedicated

hardware accelerators...

Communication

Busses, networks,

dedicated devices...

antennas...

Receiving inputs

Memories, sensors,

Providing results

Memories, display

actuators, wires...


Components (2)

Storage• Read-Only Memory: ROM• Programmable Memory: PROM, EPROM, EEPROM…• Random Access Memory: SRAM, DRAM…• Mass storage

— Hard Disk Drives (HDD, magnetic)— Solid State Drives (SSD, flash) non-volatile memories

Input/Output• Port or bus controller: UART, GPIO, USB, Ethernet…• Graphics controller, keyboard controller• Wireless card, sound card…• Analogue devices: (RF, microphone, sensors…)


https://perso.telecom-paris.fr/pacalet/CompArch/

https://gitlab.eurecom.fr/renaud.pacalet/comparch

Components (3)

Computing• Integer unit, CPU, coprocessor, Floating Point Unit (FPU),…• Hardware accelerators

Communication (internal)• Buses & bridges, Network on Chip (NoC)• Point-to-point links: interrupt, on chip accelerator…

Performance improvement units:• Direct Memory Access (DMA) controller• Interrupt controller• Caches, Memory Management Unit (MMU)


Interconnection, example architecture

CoP

InterruptHandler

CA

CH

E

MM

U

DMA

UART

(scanner,...)

(printer,...)

HardwareAccelerator

GraphicBoard (screen,...)

Busses

Bridges

(mouse, modem,...)ROM

CPU

CPU

RAM

USB

PIO


Integration…

CHIP

RAM

DMA

UART

PIO

USB

DeviceBoard

Hard.Accel.

CPU

CoP

CA

CH

E

MM

U

IRQ

ROM

RAM

DMA

UART

PIO

USB

DeviceBoard

Hard.Accel.

CPU

CoP

CA

CH

E

MM

U

IRQ

CHIP

ROM

RAM

DMA

UART

PIO

USB

DeviceBoard

Hard.Accel.

CPU

CoP

CA

CH

E

MM

U

IRQ

CHIP

ROM

RAM

DMA

UART

PIO

USB

DeviceBoard

Hard.Accel.

CPU

CoP

CA

CH

E

MM

U

IRQ

CHIP

ROMCtrl.

Mem.


PC architectures: variants, evolutionsMonitor

Keyboard Floppy disk drive

Hard disk drive

Hard disk

controller

Floppy disk

controller

Keyboard controller

Video controllerMemoryCPU

Bus

Single bus (ISA), 80’s��Memory bus

CPU PCI bridge

SCSI controller

SCSI disk

Network controller

Video controller

Printer controller

Sound card ModemISA

bridge

SCSI scanner

Main memorySCSI

bus

PCI bus

ISA bus

cache

Multi-bus (ISA + PCI), 90’s

(a) (b)

CPU

Shared memory

Bus

CPU CPU CPU

Local memories

CPU

Shared memory

Bus

CPU CPU CPU

Multiprocessor

ISA: Industry Standard ArchitecturePCI: Peripheral ComponentInterconnect


PC architectures: variants, evolutions

VIA chipset KT400 for AMD Athlon XP, 2000’s



PT894 for Intel Pentium IV, 2000’s

P55-iCore7, 2010’sDMI: Direct Media Interface



http://www.gaisler.com/

Leon Sparc by Cobham Gaisler AB



Leon Sparc by Cobham Gaisler AB21 / 64 March 2, 2021 Telecom Paris Institut Mines-Telecom Computer Architecture — Introduction

http://www.gaisler.com/

Multilevel architecture, principles

Level 0, language 𝐿0

Level 𝑛, language 𝐿𝑛

Level 𝑖, language 𝐿𝑖

Level 1, language 𝐿1

𝐿𝑖 sentences translated to 𝐿𝑖−1• Interpreted (on-the-fly)• Compiled (off-line)• Interpreted + Just-in-Time

CompiledSame principle, many use cases• Software (programming

languages)• Networking (protocols)


Multilevel computer architecture

CompArch

compiler or virtual machine

Lev. 3: Operating system

Lev. 4: Assembly language

Problem−oriented languageLev. 5:

Lev. 2: Instruction set architecture

partial interpretation

interpretation or direct execution

translation: assembler

hardware

Lev. 0: Digital

Lev. 1: Microarchitecture

user

Hardware

System software

Environment

Application software


Hardware

CPU, Instruction Set implementationMicro−Architecture

Digital

Physical

System

Micro−electronics, transistors

Logic gates, combinatorial logic

Memory elements, memories

Computer, CPU + peripherals


Micro-electronics, logic gates

Micro-electronics: transistors. Signal: voltageDigital: logic gates. Signal: boolean (0,1)


Combinatorial glue logic

input output

propagation

time

input events

output effects

Input changes immediately start propagating to outputsPropagation delays unavoidable (up to now)Frequent need to wait until signals stabilize


Combinatorial circuits (1)

Data processing example: 1-bit adder (with carry)

A B Carry in Carry out Sum0 0 0 0 00 0 1 0 10 1 0 0 10 1 1 1 01 0 0 0 11 0 1 1 01 1 0 1 01 1 1 1 1

Carry in

AB

Sum

Carry out



Signal selection: datapath controlMultiplexor: select one input among several

F

D0

D1

D2

D3

D4

D5

D6

D7

A B C

A A B CB C

form, 𝑛 bits)

𝐷0

𝐷𝑘

𝐷2𝑛−1

𝑘 (binary



Select by index: control signals, addressesDecoder: assert one output among several

D0

D1

D2

D3

D4

D5

D6

D7

A

C

B

A

B

C

B

C

A

form, 𝑛 bits)

𝐷0

𝐷𝑘

𝐷2𝑛−1

𝑘 (binary


Combinatorial circuits (4)ALU: Arithmetic and Logical UnitInput function code selects the operation

AINVA

ENAB

Logical unit Carry in

AB

B

Enable lines

F0

F1

Decoder

Output

Sum

Carry out

Full adder

A + B

ENB

code op

ALU A op B

B

A


Storage (1)

D Q

CK

(a)

D Q

CK

(b)

D Q

CK

(c)

D Q

(d)

CK

Memory elementsLatches and D-flip-flop: Q takes the value of D(a) D latches. When enable is 1(b) D latches. When enable is 0(c) D-flip-flops. On rising edge(d) D-flip-flops. On falling edge

D-flip-flops: input signal must be stable around clock edge


Synchronous hardware

Clock driven: outputs production driven by clock signalExample: falling edge triggered D-flip-flopsToggling values considered only at clock falling edges (regularity)Simplified design: propagation delay can be ignored but……D-flip-flops based ⇒ input signals stable

Input events

Outputs

Outputs

Inputs

clock

clock

time

Falling edges


Storage (2)

Multi-bit registersSRAM (Static RAM)

2 bits D-flip-flop register: 8 bits D-flip-flop register:


Storage (3)

Memory arrays𝑘-bits cells ⇒ 𝑘-bits data bus2𝑛 cells ⇒ 𝑛-bits addressesMemory size: 2𝑛×𝑘 bits

deco

der

data bus

address bus

Cells

0

1

2

3

4

5

6

7

8

9

10

11

0

1

2

3

4

5

6

7

0

1

2

3

4

5

Example: three 96−bits memory orgnisations

6×16-bits cells𝑛 = 4, 𝑘 = 8 𝑛 = 3, 𝑘 = 12

8×12-bits cells12×8-bits cells𝑛 = 3, 𝑘 = 16


Storage (4)Enable/disable register write, 2 approaches:

...but conbinatorial logicon clocks⇒ skews...

write enable

in outmultiplexer (mux)

inwrite enable

out

beware propagation delays

CLK CLK

Register access: asynchronous read, synchronous write

CLK CLK

decoder

address

output value

0 1 0

ci dia0

0

a0

a0

a1

1

ci

di

ai cibi di

write enable

input value

a1

R1

R1

R0 R0


Storage (5)

Data in

Write gate

I0

I1

I2

QD

CKWord 0

Word 1

Word 2

Word 3

O1

O2

O3

CS

RD

OE

Word 0 select line

Word 1 select line

Word 2 select line

CS • RD

A0

A1

Output enable = CS • RD • OE

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK

QD

CK D0

A0A1A2A3A4A5A6A7A8A9

A10A11A12A13A14A15A16A17A18

D1D2D3D4D5D6D7

WE

(a)

512K 3 8 Memory

chip

(4 Mbit)

CS OE

A0A1A2A3A4A5A6A7A8A9

A10

RAS

CAS

D

WE

(b)

4096K 3 1 Memory

chip

(4 Mbit)

CS OE


From combinatorial to synchronous

Registers insertion for component synchronization

c

d

a cb

x

x

y

y

z

z

F(.) F(.)

F(.)

F(x)

F(x)

F(z)

F(y) F(x) F(y) F(w)

F(x) F(y) F(z)

F(x) F(y)

F(y) x y w

dcba

cba

a

b

c

x y z a

b

c

x y z w

a

b


Exercise on synchronous logic

? Exercise #1: Complete the waveforms

5

14

o

x

s

b

a

i

t

1 1 0

a

b

s

i

k

o

10

11

0

7

0

8

12

1

1

9 2

0

6

1

5

9

0

8

2

8

5

6

9

5

5

3

+ m

u

x

+

a

b

s

0

10

o

i ?

k

m

u

x

m

u

x+

1

a

b

i

s

x

o0

1

0

t


Micro-architecture: CPU

CPU instruction: binary code of “elementary” CPU operationRunning program = set of (ordered) instructions stored in memory

⇒ CPU loads instructions from memory, decodes and executes

CPU

Program counter

Register file

external}

devices

{on chip

Memory

ports

addresses

unmapped

Coprocessors, other CPU features

Control wires (interrupts,...)

Computing units

on boardregister

Instruction

Control logic INTERCONNECT

asso

ciat

es a

ddre

sses

to l

oca

tions

to execute instruction

Code of currentinstruction

(drive multiplexors, ...)

instruction address in memory

(optimisation)

(UAL,...)

fast access to local data


CPU communication

Address bus (𝑛 bits) specify location, data bus convey data⇒ Address space = {0,…,2𝑛−1}Dedicated wires for interrupt signalling, coprocessors, surrounding hardware of multi-coreCPU….Note: CPU also export information about its internal state

Deviceinterfaceregisters

Unmappedwriteread

CPU

0

addr

data

addresses

Memories2 −1

n

interrupt

coprocessors,...

INTERCONNECT

Control

Internals...

Memory Mapping

Address Space


CPU communication

Address bus and addressspace• Bus: n wires ⇒ 2𝑛

addresses• Memory map: component

locations in address space• Mapping logic: bus

controller, for example.

CoP

Controler

Interrupt

UART

send/receive

externalport

Dat

aB

us

Ad

dre

ssB

us

portdedicated

chip select

Bus

IU

CHIPCPU


CPU communication (example)

CoP Dat

aB

us

Ad

dre

ssB

us

portdedicated

chip select

Bus

IU

CHIPCPU

Interrupt

UART

Controler

send/receive

portexternal



CoP Dat

aB

us

Ad

dre

ssB

us

portdedicated

chip select

Bus

CHIPCPU

UART

send/receive

externalport

Controler

Interrupt

managerinterruptprogram:

IU



CoP Bu

s

portdedicated

Bus

CHIPCPU

UART

send/receive

externalport

Controler

Interrupt

IU

READUART input

chip select

Bu

sA

dd

ress

Dat

a



CoP Dat

aB

us

Ad

dre

ssB

us

portdedicated

Bus

CHIPCPU

UART

send/receive

externalport

Controler

Interrupt

program:IU

UART inputRECEIVE

chip select


CPU: instruction execution

Memories

CPU

write

Step 1: Instruction Fetch

addr

data

read

addressesUnmapped


interrupt

InstructionRegister

counter

Program

coprocessors,...

Data

Code



data

MemoriesStep 2: Execute and update

read

Return to step 1

program counter

addresses


Unmapped

CPU

write

addr

interrupt

coprocessors,...

InstructionRegister

Code

DataUP

DATE



interface

writeread

Execution can cause other accesses

data

addr

addressesUnmapped

Device

registers

MemoriesCPU

interrupt

UPDATE

Data

InstructionRegister

coprocessors,... Code



program counter is set to specific interrupt service routine

addrCPU

writeread

data

Memories

Interrupt lines interrupt running programs :

addresses

Device

registersinterface

Unmapped

interrupt

Register

Codecounter

Data

Instruction

coprocessors,...

Program

Changed


What an instruction can do

Read or write in address spaceRead or write in CPU data registersWrite Program Counter (PC)Write CPU configuration registersCompute values and store them somewhereDrive coprocessor…Chain several such action, that is…execute a microprogram…


Instruction Set Architecture

An ISA is a well-define hardware/software interface:• An instruction set with a functional definition and a user manual• No guarantees regarding implementation and performance

History and practice: various instruction sets• Realisation: choices between hardware and software

Evolution:

Simple instructions

Hardware Hardware

Complex instructions

Software

Complex Programs

Software

Simple Programs


Instruction set: evolution

CISC

Complex Instruction Set ComputerReduced number of simple instructions ⇒ long complex programsComplex instructions in hardware ⇒ increased performanceMicroprogrammed instructions ⇒ easier to extendSimple programs with complex instructions• lower CPU clock frequency or…• …more CPU clock cycles per instruction


Instruction set: last 25 years trend

RISC

Reduced Instruction Set ComputerSimple instructions ⇒ one per clock cycleSimple instructions ⇒ higher clock frequencySimpler compilersElementary instructions faster than microprogramsPipeline: parallel execution of instructionsNote: RISC and CISC are sometimes combined (ex: Intel)


Instruction Set / Micro-architecture

Pipelined architecture (typical of RISC)• Split instruction into elementary steps• Each step requires different hardware components• Parallel execution of instructions: one step per clock cycle for each instruction in pipeline• All hardware components are used simultaneously• Goal: start one new instruction per clock cycle


ISA: Instruction Set Architecture

Instruction examples• Simple ones

— Load memory word in register: ld R1 Addr— Store register in memory: st R1 Addr— Add registers: add R1 R2 R3

• Complex ones (combining memory access and logic)— Load memory word and add to register: ladd R1 Addr

Instruction format: binary• Variable or fixed size• Structured format: fields• Duration: fixed or variable (1 cycle or more)

An ISA is often provided with an assembly language


Assembly language

Binary code: 0100110110001010111010001010011101010Assembly language, a more readable format:• CPU instructions in textual form with parameters:

— ld R1 R2: load into R1, from address in R2.— jmp #0820: jump at address #0820.— beq R1 R2 #0840 branch at #0840 if R1 and R2 have the same content

• Macros and tools for easier programming:— Pseudo-instructions (abbreviate sequence of real instructions)— Labels:

label1: ld R1 R2;...jmp label1

Assembler (assembly language compiler): reads assembly language and produce binarycode.


Running software: ways to invoke code

Succession of instructions from bootJumps allow to alter the sequence of instructions in memoryCalling subroutines (with return)

rs

PC

a1

aib1

bmai+1

an

call f: js b1

save ai+1

return: rs

restore ai+1

js b1ai

an

bm

tim

e

Memory

P

fb1

a1


Running software: ways to invoke code

Interrupt ReQuest (IRQ) signalIRQ cause CPU to jump to Interrupt Service Routine (ISR)PC set to address of ISRUsually return to main code at end of ISR

PCsignal

tim

ea1

aib1

bmai+1

an

IRQ

signal

control

PC

CPU

b1

IRQ

0

1

0

0

0

0

save context

go to ISR

P

ISR

Memory

an

b1

bm

a1

rireturn: ri

restore context

mux


Software: structuring

Structuring software:• Operating System (OS): services and task management

— Load and execute application software— When multi-task, control task access to resource (CPU, memory…)— Libraries: available code, system calls

• Application software— Focus on specific work, do not manage low level details— Use system calls (libraries)— Is loaded and cleanly terminated by OS

applicationsystem

Memory

CPU


OS: executing programs

From software saved in binary file on disk to process running in memory1 Find free memory for software code2 Load code in memory3 Identify and locate external resource required by software

— External addresses— External code to load (libraries, system code…)

4 Update code to reflect addresses w.r.t. points 1 and 35 Set PC to entry point of code6 At software termination restore clean global state

� Files containing executable code must be packaged with associated meta-information� Dedicated binary file formats (e.g. ELF)


Boot sequence

Run code at boot address (hard-wired)• Bare metal code (embedded systems, no-OS systems…)• Operating system (GNU/Linux, Windows, Android…)

— Installs system services: ISRs, device drivers…— If multi-task, installs scheduler— Run system processes (daemons…, depends on system)

• Example: PC boot sequence.— First instructions: hardware initialisation (BIOS)— System initialisation: load Master Boot Record (MBR) from disk— Execute MBR (first system instructions), usually jump to initialisation code located elsewhere— Initialisation code loads next stage— Next stage loads next stage…— System reaches stable state— Interactive programs wait for user inputs


Course Content

Basics of binary representations and data typesISA and MIPS assembly languageMIPS architecture (mono-cycle, multi-cycle, pipeline)Pipelining, hazardsCaches and memoriesHardware support for OS (supervisor mode, atomic operations, MMU, interrupt controller,timer, DMA…)Labs: MIPS core simulationARM mini-conferences (ARM CPU architectures, design, validation)


Summary

Questions?


Further Reading I

D.A. Patterson and L. HennessyComputer Organisation & DesignMorgan Kaufmann, USA

D.A. Patterson and L. HennessyComputer Architecture, A Quantitative approachMorgan Kaufmann, USA

And a lot more in the library…


computer architecture -...

Documents