Computer Architecture

----A Quantitative Approach

College of Computer Science & Technology

Jiang, Xiaohong
2023.fall

1
 Instructor & TA
Instructor ： Jiang Xiaohong
 Office : Room520, Bld. of CaoGuangBiao,
 Mobile(short): 529114
 Email: [email protected]
 Homepage:
http://mypage.zju.edu.cn/jiangxh

TA ： Lan Haoxuan （兰浩轩）

Course Website ：
 course.zju.edu.cn

2023/9/19 1.2
2
 Why we learn Computer Architecture?
 Hardware Course Series

并行处理 高级计算机体系结构 分布式系统

Parallel Processing Advanced CA Distributed System

存储技术基础 接口设计 计算机体系结构 嵌入式系统 多核程序设计

Storage Multicore
Interface Computer Embedded
Technology Programming
Fundations
Design Architecture System

计算机组成
Computer Organization

逻辑与计算机设计基础
Logic and Computer Design Fundamentals

电子线路 模拟线路
Electronic Circuits Simulation Circuis

2023/9/19 1.3
3
 Why we learn Computer Architecture?
嵌入式系统
编译系统 计算机网络

操作系统 计算机体系结构 汇编与接口 大学三年级

计算机组成
数据库系统
大学二年级
高级数据结构与算法分析
数字逻辑设计基础

数据结构基础

计算机系统概论
大学一年级
C 语言程序设计 大学计算机基础 通识课程

2023/9/19 1.4
4
 Different from Organization
Applications

O perating System

 Part & Whole Com piler Firm w are

 Bottom up & Top down Instruction Set Architecture

 How & Why Instruction Set Processor

Datapath & Control

I/O System

Digital Design
Circuit Design

Layout

5
 Course Objectives
The objective of this course
 systemically learn the fundamental concepts and design
approaches of computer architecture using quantitative
approaches from the view of the whole computer system.
 Learn the ideas and approaches to improve the performance of
computer system via exploring ILP, DLP, and TLP.
 Grasp the hardware design tools and environment, design and
implement the hardware with Verilog language in vivado
environment on FPGA board.

2023/9/19 1.6
6
 Textbook
David A. Patterson, John L. Hennessy,

《 Computer Architecture
– A Quantitative Approach 》

6th Edition. July , 2019.

China Machine Press,
ISBN: 978-7-111-63110-1

2023/9/19 1.7
7
 John L. Hennessy （ Stanford ）
 Former President of Stanford University during
2000 – 2016 （ 17 billion ）
 Current Alphabet Chairman
 "Godfather of Silicon Valley,“

 In 1981, Hennessy initiated a project at Stanford

that focused on a simpler computer architecture
known as RISC. During a sabbatical leave in
1984-85 he cofounded MIPS Computer
Systems, now known as MIPS Technologies,
which specializes in the production of
microprocessors SPARC.
https://
engineering.stanford.edu/  Received Eckert-Mauchly Award in 2001
people/john-hennessy  Received Turing Award in 2017
8
 David A. Patterson （ UC Berkeley ）
 UC Berkeley （ 1976 – 2016 ）
 Currently Google TPU

 He led the design and implementation

of RISC I (the foundation of the
SPARC architecture )
 Inventor of RAID
 involved in the Network of
Workstations (NOW) project
 Research Accelerator for Multiple
Processors (RAMP)
 Received ACM Eckert-Mauchly Award https://people.eecs.berkeley.edu/~pattrsn/
in ISCA 2008
 Received Turing Award in 2017
9
 Text book evolution
1-3 edition 4-5th edtion 6th edtion

• MIPS • MIPS / ARM • RISC V

• ILP （ Instruction Level • ILP （ Instruction Level • ILP （ Instruction Level

Parallelism ） Parallelism ） Parallelism ）
DLP (Data Level Parallelism ） DLP (Data Level
TLP （ Thread Level Parallelism ）
Parallelism ） TLP （ Thread Level
• Cost / Performance Parallelism ）
• Cost / Performance RLP （ Request Level
dependability / power Parallelism ）

• PC • dependability
• PC Cost / Performance/power
embedded system
Server • PC
Personal mobile device
2023/9/19 warehouse-scale computers 1.10
10
 Updated Course Contents
教材： 6th Edition
教材： 5th Edition Contents in 2021 备注
Contents in 2020 备注 Ch1 Fundamentals of computer
1.1 ~ 1.13
design
Ch1 Fundamentals of computer
design Ch2 Memory Hierarchy Design 2.1 ~ 2.9, app B3
App C2-C6 , 3.1 ~
AppA Instruction Set Principles MIPS
3.14
AppC Pipelining: Basic and Pipeline Ch3 ILP and its expoloitation Branch prediction
Intermediate Concepts CPU Dynamic scheduling
Speculation
AppB Review of Memory
Hierarchy Ch4 DLP in Vector, SIMD, 4.1~ 4.9
GPU Vector. SIMD. GPU
Improve
Ch2 Memory Hierarchy Design
cache perf. 5.1~ 5.11
Multiprocessor
AppD Storage Systems Ch5 Thread Level Parallelism
Cache coherence
AppI Basic concepts of Synchronization
Multiprocessor Ch6 Warehouse-scale computer
选讲
to explore RLP & DLP
2023/9/19 1.11
11 Ch7 Domain-specific
 2018 interview
For pioneering a systematic, quantitative
approach to the design and evaluation of
computer architecture with enduring impact
on the microprocessor industry.

This book is essential reading for those

tasked with leading any complex
enterprise in the academic, not-for-profit,
or for-profit sector.

https://baijiahao.baidu.com/s?id=1607140815624945357&wfr=spider&for=pc
2023/9/19 1.12
12
 How ?

Concepts, Ideas and Principles

Quantitative approaches
Hit the problem and right way to solve problem

 As a man sows, so he shall reap.

一分耕耘一分收获

2023/9/19 1.13
13
 Grading Policy:
 16-20%: written homework
28%
 8-12%: pop quiz 2-3 times
 32%: Lab assignments 60% ( <=60 )
 5-10%: Bonus
 lab grade = report (50%) + check(50%)

 40%: Final exam

 (close-book test with one A4 memo)

 Final grade = 40%(Final exam ） + 60% (other <=100)

2023/9/19 1.14
14
 Lectures vs. Labs
Week Theory(lectures) Labs %
1 Ch0: Course Introduction. Lab1: Enhance CPU with forwarding and predict-NT 6
2 Ch1-1: CA design Lifecycle
3 Ch1-2: trends of CA engineering
4 Ch2-1: cache optimization Lab2: Adding exception to CPU in Lab1 4
5 Ch2-2: VM and VA
6 Ch3-0: ILP--supporting multicycle OP Lab3: implement a 2-way set associative cache 3
7 Ch3-1: scoreboard
8 Ch3-2: Tomasulo Lab4: adding cache to CPU of Lab2 4
9 Ch3-3: hardware-based speculation
10 Ch3-4: multiple issue processor Lab5 ： expending CPU to support multicycle OP 7
11 Ch4-1: DLP—vector processor
12 Ch4-2: SIMP ， GPU
Lab6 ： implement dynamic scheduling:
13 Ch5-1: TLP-multiprocessor 8
scoreboard/Tomasulo
14 Ch5-2: cache coherence
15 Ch5-3: synchronization
16 Review

2023/9/19 1.15
15
 Homeworks (16-20%)
Total 4-5 times, once per chapter
Submission deadline will be normally one week after
assigned, and will be announced on course
website.
For doing homework, discussion is greatly
encouraged, but every student is required to Do and
Submit the homework individually on time.
Submission Naming rule
 StID_name_hw1.doc

2023/9/19 1.16
16
 Lab assignments
Lab1----Implement pipelined CPU with forwarding paths and
prediction-not-taken supporting RISC V 32i instructions.
Lab2----Implement Interruption and Exception on CPU of Lab
1).
Lab3----Implement a 2-way associative cache
Lab4----Adding the cache into the pipelined CPU in Lab 2).
Lab5----Expend the pipelined CPU to support multi-cycle
operations: integer multiplier, integer divider / remainder
operation, issue in order and complete out of order, detecting
pipeline hazards ( WAW, RAW)
Lab6----Implement the Dynamic Scheduling (Scoreboard or

2023/9/19
Tomasulo), implement a dynamic scheduling pipelined CPU.
1.17
17
 Labs ( 32%)
 Grading (report 50% + check 50%)/per lab
 Lab1 3 weeks, 6% personally
 Lab2 2 weeks, 4% personally
 Lab3 1 weeks, 3% personally
 Lab4 2 weeks, 4% personally
 Lab5 3 weeks, 7% in group of 2 members
 Lab6 4 weeks, 8% in group of 2 members

 Top 2 labs of best performance(fastest, lower cost) will get 5%

bonus
 Each lab have up to 2 times to be checked. The 2nd check will
have score <= 90.

2023/9/19 1.18
18
 Submission Policy:
 Allthe homeworks (individually) and lab reports (in
group) are required to be submitted to the course
website on time.
 Lab report and lab code are required to submit
separately using different links under different deadline.
Submission deadline will be announced on course
website.
All assignments in this course should be turned in by
the specified due date. Late assignment is NOT
accepted.

2023/9/19 1.19
19
 Honest Policy
 Be HONEST in your work!

 Found copy & be copied in the homework or lab report ，

you get ZERO for one submission and also get 10% off
in the final grade !

2023/9/19 1.20
20
 Q&A

??

2023/9/19 1.21
21