Giters

cmpark0126 / cs420

KAIST CS420: Compiler Design (2020 Spring)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

KAIST CS420: Compiler Design (2020 Spring)

Logistics

Online sessions

Due to COVID-19, we're going to conduct online sessions.

  • Online sessions will be provided via this YouTube channel.

  • You're required to watch the video, and based on the contents, to solve pop quizzes that will be posted at gg.kaist.ac.kr. The details will be announced in the issue tracker, e.g., kaist-cp#3

  • If it's difficult to understand English, please turn on the subtitles in the YouTube videos. Auto-transcribed subtitles will be shown.

Course description

Context

Compilers bridge the gap between human and machine. Human wants to easily express complex idea. On the other hand, machine understands only a few words (instructions) to be efficiently implemented in silicon. Compilers transform programs from a form suitable for human to easily express complex idea, to a form suitable for machine to efficiently execute. Since the gap between human and machine is fundamentally wide, compilers have been constructed and widely used since the beginning of the history of computing. Even, the first practical compiler predates the first practical operating systems (according to Wikipedia)!

In response to industry shifts, new compilers should be written and written again. First, human wants to express more and more complex idea, especially in the era of artificial intelligence and big data. Second, machine changes in response to physics (e.g. the ending of Dennard scaling and Moore's law) and industrial needs (e.g. Internet of Things and distributed systems). New compilers should be constructed to close the new gap between changing human and changing machine. For this reason, industrial needs for (and salary of) compiler engineers have been constantly high.

Goal

In this class, we will learn how to construct a compiler by actually building one. You are going to benefit from the provided skeleton code of a clean slate educational compiler--dubbed KECC: KAIST Educational C Compiler (think: KENS for networking or Pintos, xv6 for operating systems). We are going to discuss parsing only briefly, because the topic is assumed to be dealt with in CS322: Formal Languages and Automata. (You don't need to know parsing to take this course, though.) We will focus on translation from human-friendly form to machine-friendly form, and compiler optimizations. Specifically, we will discuss (1) how to transform a C program to an SSA-based intermediate representation (IR); (2) how to perform register promotion, static single assignment, global value numbering, and register allocation optimizations on the IR; and (3) how to transform an IR program to a RISC-V assembly program. KECC will provide a significant amount of skeleton code so that you can focus on the topic of this course.

We will also briefly discuss the recent trends of compiler construction. I see two crucial recent trends: script languages and parallelism. (1) Scripting languages like JavaScript and Python, unlike C, should be compiled (or interpreted) at run-time, and therefore, there is no clear distinction of compile- and run-time. It is a challenge in that compile time should also be optimized, but it is also an opportunity in that compile may gather and benefit from run-time information. (2) It is crucial to exploit the massive parallelism of modern applications like deep learning and high-performance computing (HPC), because they require so huge computation. Due to the complexity of workloads, their parallelism should be automatically discovered and exploited by compilers, which is a big challenge.

We will also briefly study the theory of compiler. We will focus on the correctness of compiler. In general, in what sense a compiler is correct, and how to prove it? Specifically, how to prove the correctness of KECC's transformations and optimizations? As it will turn out, this compiler correctness theory will greatly help you efficiently build your own compiler.

Resources

Tools

Make sure you're capable of using the following development tools:

  • Git: for downloading KECC and version-controlling your development. If you're not familiar with Git, walk through this tutorial.

    • IMPORTANT: you should not expose your work to others. In particular, you should not fork the upstream and push there. Please the following steps:

      • Directly clone the upstream without forking it.

        $ git clone --origin upstream https://github.com/kaist-cp/kecc-public.git
$ cd kecc-public
$ git remote -v
upstream	https://github.com/kaist-cp/kecc-public.git (fetch)
upstream	https://github.com/kaist-cp/kecc-public.git (push)

      • To get updates from the upstream, fetch and merge upstream/master.

        $ git fetch upstream
$ git merge upstream/master

    • If you want to manage your development in a Git server, please create your own private repository.

      • You may upgrade your GitHub account to "PRO", which is free of charge. Refer to the documentation

      • Set up your repository as a remote.

        $ git remote add origin git@github.com:<github-id>/kecc-public.git
$ git remote -v
origin	 git@github.com:<github-id>/kecc-public.git (fetch)
origin	 git@github.com:<github-id>/kecc-public.git (push)
upstream https://github.com/kaist-cp/kecc-public.git (fetch)
upstream https://github.com/kaist-cp/kecc-public.git (push)

      • Push to your repository.

        $ git push -u origin master

  • Rust: as the language of homework implementation. We chose Rust because its ownership type system greatly simplifies the development of large-scale system software. If you want to "opt out", you can also use FFI and implement your compiler in C/C++.

    We recommend you to read this page that describes how to study Rust.

  • Visual Studio Code (optional): for developing your homework. If you prefer other editors, you're good to go.

  • Compiler Explorer (optional): for comparing the result of KECC to LLVM IR or RISC-V assembly code. See this video for instructions.

    • NOTE: If you want to see LLVM IR code after using a specific optimization pass (e.g., mem2reg, gvn, etc.), please follow the steps:
      1. Uncheck Directives and Comments options in the filter menu at the top of the compiler window.
      2. Type your C code, and obtain LLVM IR code by compiling it using -O0 -Xclang -disable-O0-optnone -emit-llvm flags.
        • See this link for the description of each compilation flag.
      3. Copy LLVM IR code.
      4. Change the target language from C to LLVM IR and compiler from clang to opt.
      5. Paste LLVM IR code, and optimize it using -mem2reg(or -gvn) flag.

  • If you want, you'll be provided with a Linux server account. Please submit your SSH key here. You can connect to server by ssh s<student-id>@cp-service.kaist.ac.kr -p10005, e.g., ssh s20071163@cp-service.kaist.ac.kr -p10005.

    • Your private key id_ed25519 should be in ~/.ssh.

    • Add the following lines in your ~/.ssh/config:

      Host cs420
  Hostname cp-service.kaist.ac.kr
  Port 10005
  User s<student-id>

      Then you can connect to the server by ssh cs420.

    • Now you can use it as a VSCode remote server as in the video.

Prerequisites

  • It is strongly recommended that students already took courses on:

    • Mathematics (calculus MAS101 & MAS102 or discrete mathematics CS204): proposition statement and proof
    • Data structures (CS206): linked list, stack, queue
    • Systems programming (CS230): memory layout
    • Programming languages (CS320): lambda calculus, interpreter

    Without a proper understanding of these topics, you will likely struggle in this course.

  • Other recommendations which would help you in this course:

    • Basic understanding of computer architecture (CS311)
    • Programming experience in Rust

Grading & honor code

Homework (80%)

You will implement translations and optimizations on KECC. All homework submissions will be automatically graded online so that you can immediately see your score. If your compiler is correct and the generated assemblies perform comparably to those generated by gcc -O1, you're going to get A+ (if not A#) even if you miss the final exam.

Since compiler construction requires nontrivial undertaking, you're encouraged to ask questions on the homework in the issue tracker at the early stage of the semester.

Final exam (20%, 16th of June, 10:30-11:45)

The exam will evaluate your understanding of compiler theory. There will not be a midterm exam.

Attendance (?%)

You should submit a token to the Course Management website for each session. You should submit a token within 12 hours from the beginning of a session.

Honor code

Please sign KAIST School of Computing Honor Code here.

Communication

  • Course-related announcements and information will be posted on the website as well as on the GitHub issue tracker. You are expected to read all announcements within 24 hours of their being posted. It is highly recommended to watch the repository so that new announcements will automatically be delivered to you email address.

  • Ask your questions via email only if they are either confidential or personal. Otherwise, ask questions in this repository's issue tracker. Any questions failing to do so (e.g. email questions on course materials) will not be answered.

    • I'm requiring you to ask questions online first for two reasons. First, clearly writing a question is the first step to reach an answer. Second, you can benefit from questions and answers of other students.

  • Emails to the instructor or TAs should begin with "CS420:" in the subject line, followed by a brief description of the purpose of your email. The content should at least contain your name and student number. Any emails failing to do so (e.g. emails without student number) will not be answered.

About

KAIST CS420: Compiler Design (2020 Spring)