2023.12.27

Quantifying the Complexity of Superscalar Processors

1997.complexity.palacharla.isca.923.pdf的详细版

手工 参数化 分析电路的延迟

感觉这个工作综合工具就能做吧?

手工做可能会遗漏。

Introduction

in this paper we measure com- plexity as the critical path through a piece of logic, and the longest critical path through any of the pipeline stages determines the clock speed

定义“复杂度”为一块“逻辑电路的关键路径”

基本观点:IPC好测,用仿真器。 但是复杂度不好测。因为需要具体的实现电路。

本文的思路:复杂度的绝对值不好获取,但是相对变化可以测(发射宽度和发射窗口大小是自变量) 定义:

  • 窗口大小(window size):
    • the number of waiting instructions from which ready instructions are selected for issue

  • 发射宽度(issue width):
    • the number of instructions that can be issued in a cycle

2 Sources of Complexity

文章要研究的结构

  • Register rename logic
  • Wakeup logic
  • Selection logic
  • Data bypass logic

文章不研究但给出了参考文献的结构

  • Register file
    • [TODO][11] 1996 Register File Design Considerations in Dynamically Scheduled Processors
  • Caches
    • [TODO][31] 1992 An Analytical Access Time Model for On-Chip Cache Memories
    • [TODO][33] 1994 An Enhanced Access and Cycle Time Model for On-Chip Caches
  • Instruction fetch logic
    • [TODO][26] 1996 Trace Cache: a Low Latency Approach to High Bandwidth Instruction Fetching

3 Methodology

有两类处理器模型baseline模型和reservation station模型。

  • [TODO] reorder buffer只有reservation station有???
  • [TODO] P7: reorder buffer和reservation station不同?但图2中没画reservation station啊?

很复杂,但是这里讲的不够基础,跳跃性很强。 只用看这里的结论就好啦。 之后有兴趣应该去看更低层的电路书籍。

4.1 Logic delays

逻辑延迟随feature size正比下降

4.2 Wire delays

TLDR:线延迟和线的电阻$R_{metal}*L$、线的电感$C_{metal}*L$成正比,和线长L没关系(我:光速下传输延迟这么短的线可以忽略)。 $R_{metal}$和$C_{metal}$与feature size(S,即长度L)成反比。

所以尽管线变短,总的电阻和电感也不会变。

我的总结

如果不该变设计,只是feature size减小(整体等比缩小,电路和逻辑门都等比缩小), 则线延迟会变成主要延迟。

5 Complexity Analysis

  • Figure 5看不懂
    • 去看[34] 1996 《MIPS R10000 Superscalar Microprocessor》
      • 和[34]的Figure 4有点相似?Amp是啥?原文中有提到sense amplifiers
        • 去看2015《the art of electronics》搜索sense amp。搜到了14.4.3 Static RAM。
          • 从10.4.1 Devices with memory: flip-flops即触发器的基础开始看起