Fundamentals of Computer Design

Introduction

Von Neumann Structure

Classes of Computers

• Desktop computers
  PC: Personal Computers
• Servers computers
  更强大的处理速度，容量（用于冗余备份）
• Embedded computers
  不能随意安装第三方应用的，与系统一体，称为嵌入式（不太符合国情x
• Personal Mobile Devices
  如手机，iPad
• Supercomputer

Classed by Flynn
按照指令流和数据流进行分类

• SISD
  单指令流单数据流，如早期的单核 PC
• SIMD
  一条指令有多条数据流动（如向量数据），方便做流水线
• MISD
  多指令流单数据流，并不实际存在
• MIMD
  多指令流多数据流

Performance

• Alogrithm
• Programming language, compiler, architecture
• Processor and memory system
• I/O system (including OS)

Summary

According to the process of using data, computers are developing in three fields:

• speed up processing (parallel)
• speed up transmission (accuracy)
• Increase storage capacity and speed up storage (reliability)

Performance

这里有很多因素会影响性能：体系结构，硬件实现，编译器，OS...

We need to be able to define a measure of performance.

• Single users on a PC -> a minimization of response time
• Large data -> a maximization of throughput

为了衡量性能，我们有响应时间和吞吐量两个指标：

• Latency (Response time 响应时间)
  一个事件开始到结束的时间
• Throughput (bandwidth 带宽)
  给定时间范围内完成了多少的工作量

这部分可见计组笔记

The main goal of architecture improvement is to improve the performance of the system.

Technology Trend

The improvement of computer architecture

• Improvement of input / output
• The development of memory organization structure
• Two directions of instruction set development
  • CISC / RISC
• Parallel processing technology
  不同层次、粒度的并行

Quantitative approaches

CPU Performance

• CPU 执行时间 = CPU 时钟周期数 * CPU 时钟周期时间 = CPU 时钟周期数 / CPU 时钟频率
• IC：Instruction Count，指令数
• CPI：Cycle Per Instruction，每条指令的时钟周期数
  • 由 CPU 硬件决定
  • 不同的指令也会有不同的 CPI，平均 CPI 取决于指令的组合方式
  • CPI = CPU 时钟周期数 / IC
  • CPU 执行时间 = IC * CPI / CPU 时钟频率

Amdahl's Law

Amdahl's Law: the performance improvement to be gained from using some faster mode of execution is limited by the fraction of the time the faster mode can be used.
当提升系统性能时，有多大的收益受限于被提升的部分所占的运行时间比例

$T_{improved}={\displaystyle \frac{T_{affected}}{\text{improvement factor}}}+T_{unaffected}$

Make the common case fast!

也被用来分析可行性

• 加速比

  $$\begin{array}{rl}\text{Speedup}& ={\displaystyle \frac{{\text{Performance for entire task}}_{\text{using Enhancement}}}{{\text{Performance for entire task}}_{\text{without Enhancement}}}}\\ & ={\displaystyle \frac{{\text{Total Execution Time}}_{\text{without Enhancement}}}{{\text{Total Execution Time}}_{\text{using Enhancement}}}}\end{array}$$

  加速比 Sp = 改进后的性能 / 改进前的性能 = 改进前的时间 / 改进后的时间

• 执行时间
  $T_{new}=T_{old}\times ((1-f)+{\displaystyle \frac{f}{Sp}})$
  $f$ 指改进的部分所占的比例

• $S{p}_{overall}={\displaystyle \frac{T_{old}}{T_{new}}}={\displaystyle \frac{1}{(1-f)+{\displaystyle \frac{f}{Sp}}}}$
  • 其中 $Sp$ 为被优化部分的加速比，$S{p}_{\text{overall}}$ 为整体加速比，$f$ 为被优化部分所占的运行时间比例

Great Architecture Ideas

• 摩尔定律
  • 每过 18-24 个月，集成电路的晶体管数量将增加一倍
• 使用抽象来简化设计
• 让最常见的情况更快
• 通过并行来提高性能
• 由很多级别的并行，比如指令集并行、进程并行等
• 通过流水线来提高性能
  • 将任务分为多段，让多个任务的不同阶段同时进行
  • 通常用来提高指令吞吐量
• 通过预测来提高性能
• 使用层次化的内存
  • 让最常访问的数据在更高层级，访问更快

ISA

• Instruction Set Architecture

Instruction Set Design Issues

• Where are operands stored?
  registers, memory, stack, accumulator
• How many explicit operands are there? (Classification of ISAs)
  0, 1, 2, or 3
• How is the operand location specified? (Addressing Modes)
  register, immediate, indirect, ...
• What type & size of operands are supported? (Data Representation)
  byte, int, float, double, string, vector, ...
• What operations are supported? (Types of Instructions)
  add, sub, mul, move, compare, ...

Basic Principles

• Compatibility
• Versatility
• High efficiency
• Security

ISA Classification Basis

这里主要指的是从哪里取数，存到哪里以及计算的规则。

• stack First operand removed from second op replaced by the result.
• accumulator
  • One implicit operand: the accumulator; one explicit operand: mem location
  • Accumulator is both an implicit input operand and a result
• register
  • Register-memory architecture
    任何指令都可以访存
  • Load-store architecture
    只有 load/store 的时候才能访存，其他时候都是基于寄存器操作

GPR Classification

A+B

More: try to do with $D=A\ast B-(A+C\ast B)$

GPR 速度快，但是 GPR 太多也会有资源的浪费和性能下降（如寻找对应的寄存器）

评论