Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
YJIT: Dive into Ruby's JIT compiler written in ...
Search
Takashi Kokubun
September 23, 2022
Programming
2
2.2k
YJIT: Dive into Ruby's JIT compiler written in Rust / Rust.Tokyo 2022
Rust.Tokyo 2022
Takashi Kokubun
September 23, 2022
Tweet
Share
More Decks by Takashi Kokubun
See All by Takashi Kokubun
ZJIT: The Ruby 4 JIT Compiler / Ruby Release 30th Anniversary Party
k0kubun
1
320
ZJIT: The Future of Ruby Performance / San Francisco Ruby Conference 2025
k0kubun
1
61
ZJIT: Building a New JIT Compiler for Ruby / REBASE 2025
k0kubun
0
63
Deoptimization: How YJIT Speeds Up Ruby by Slowing Down / RubyKaigi 2025
k0kubun
2
3.7k
YJIT Makes Rails 1.7x faster / RubyKaigi 2024
k0kubun
7
15k
Ruby JIT Hacking Guide / RubyKaigi 2023
k0kubun
2
10k
Towards Ruby 4 JIT / RubyKaigi 2022
k0kubun
3
12k
Optimizing Production Performance with MRI JIT / RubyConf 2021
k0kubun
1
510
Why Ruby's JIT was slow / RubyKaigi Takeout 2021
k0kubun
3
2k
Other Decks in Programming
See All in Programming
JETLS.jl ─ A New Language Server for Julia
abap34
2
480
生成AIを利用するだけでなく、投資できる組織へ
pospome
2
440
AIエージェント、”どう作るか”で差は出るか? / AI Agents: Does the "How" Make a Difference?
rkaga
3
1k
20251212 AI 時代的 Legacy Code 營救術 2025 WebConf
mouson
0
250
メルカリのリーダビリティチームが取り組む、AI時代のスケーラブルな品質文化
cloverrose
2
460
ELYZA_Findy AI Engineering Summit登壇資料_AIコーディング時代に「ちゃんと」やること_toB LLMプロダクト開発舞台裏_20251216
elyza
2
1k
The Art of Re-Architecture - Droidcon India 2025
siddroid
0
160
[AI Engineering Summit Tokyo 2025] LLMは計画業務のゲームチェンジャーか? 最適化業務における活⽤の可能性と限界
terryu16
2
300
実はマルチモーダルだった。ブラウザの組み込みAI🧠でWebの未来を感じてみよう #jsfes #gemini
n0bisuke2
3
1.4k
AI Agent の開発と運用を支える Durable Execution #AgentsInProd
izumin5210
7
1.7k
Kotlin Multiplatform Meetup - Compose Multiplatform 외부 의존성 아키텍처 설계부터 운영까지
wisemuji
0
170
例外処理とどう使い分ける?Result型を使ったエラー設計 #burikaigi
kajitack
16
5.3k
Featured
See All Featured
Chrome DevTools: State of the Union 2024 - Debugging React & Beyond
addyosmani
9
1k
Reality Check: Gamification 10 Years Later
codingconduct
0
2k
From Legacy to Launchpad: Building Startup-Ready Communities
dugsong
0
120
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
kimpetersen
PRO
0
210
Digital Ethics as a Driver of Design Innovation
axbom
PRO
0
140
Visualization
eitanlees
150
16k
Ten Tips & Tricks for a 🌱 transition
stuffmc
0
49
Building Flexible Design Systems
yeseniaperezcruz
330
40k
The Pragmatic Product Professional
lauravandoore
37
7.1k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
49
3.3k
Distributed Sagas: A Protocol for Coordinating Microservices
caitiem20
333
22k
Intergalactic Javascript Robots from Outer Space
tanoku
273
27k
Transcript
YJIT: Dive into Ruby's JIT compiler written in Rust @k0kubun
/ Rust.Tokyo 2022
Me • @k0kubun • Shopify ◦ YJIT team • Ruby
committer ◦ MJIT maintainer
What’s YJIT?
None
None
None
None
How does YJIT work?
How YJIT works Ruby code
How YJIT works 1 + 2 Parse Ruby code Abstract
Syntax Tree
How YJIT works 1 + 2 Parse Ruby code Abstract
Syntax Tree putobject 1 putobject 2 opt_plus leave Compile Bytecode
How YJIT works 1 + 2 Parse Ruby code Abstract
Syntax Tree putobject 1 putobject 2 opt_plus leave Compile JIT Bytecode Machine code
How YJIT works putobject 1 putobject 2 opt_plus leave JIT
? Machine code Bytecode
How YJIT works: Ruby 3.1 putobject 1 putobject 2 opt_plus
leave x86_64 Codegen Machine code Bytecode
How YJIT works: Ruby 3.2 putobject 1 putobject 2 opt_plus
leave Machine code Bytecode
How YJIT works: Ruby 3.2 putobject 1 putobject 2 opt_plus
leave Machine code Bytecode
Lazy Basic Block Versioning
Lazily compile basic blocks
getlocal a getlocal b opt_plus setlocal c getlocal c putobject
1 opt_gt branchunless getlocal a leave getlocal b leave Lazily compile basic blocks
Lazily compile basic blocks
getlocal a getlocal b opt_plus setlocal c getlocal c putobject
1 opt_gt branchunless Branch stub Branch stub Lazily compile basic blocks
getlocal a getlocal b opt_plus setlocal c getlocal c putobject
1 opt_gt branchunless getlocal a leave Branch stub Lazily compile basic blocks
getlocal a getlocal b opt_plus setlocal c getlocal c putobject
1 opt_gt branchunless getlocal a leave Branch stub Lazily compile basic blocks
getlocal a getlocal b opt_plus setlocal c getlocal c putobject
1 opt_gt branchunless getlocal a leave getlocal b leave Lazily compile basic blocks
Why lazy compilation? 1. Better code locality a. Only compile
used paths b. Related code is put together
Why lazy compilation? 1. Better code locality a. Only compile
used paths b. Related code is put together 2. More type information
Type Profiling
Type Profiling getlocal a getlocal b opt_plus
Type Profiling getlocal a getlocal b opt_plus
Type Profiling getlocal a getlocal b jmp regenerate
Type Profiling getlocal a getlocal b jmp regenerate
Type Profiling getlocal a getlocal b jmp regenerate a: 1
b: 2
Type Profiling getlocal a getlocal b opt_plus (int) setlocal c
getlocal c putobject 1 opt_gt branchunless
Passes • CodeGen -> Split -> Alloc Regs
Passes • CodeGen -> Split -> Alloc Regs
IR
IR Split
IR Split Alloc Regs
Rust Challenges
Clang and Bindgen Clang dependency or Per-architecture codegen?
Mutable Borrow
Next steps • Code GC • Register allocation • Method
inlining
Conclusion • We reviewed the architecture of YJIT • Rust
has been useful for transforming IR