ZJIT: The Future of Ruby Performance / San Francisco Ruby Conference 2025

https://sfruby.com/

Transcript

1. San Francisco Ruby Conference 2025 ZJIT: The Future of Ruby Performance Takashi Kokubun / @k0kubun
2. self • Takashi Kokubun (@k0kubun) • Shopify: Ruby JIT team • Ruby committer: MJIT, RJIT, YJIT, ZJIT
3. YJIT
4. YJIT • Ruby 3.1+: ruby --yjit • Production-ready JIT compiler • Enabled by default on Rails 7.2+
5. https://speed.ruby-lang.org/
6. https://railsatscale.com/2025-01-10-yjit-3-4-even-faster-and-more-memory-e ff i cient/
7. ZJIT
8. ZJIT • Ruby 4.0+: ruby --zjit • Experimental JIT compiler • To be productionized at Ruby 4.1
9. Why ZJIT? • Unblock cross-instruction optimizations • Less incremental, larger comiplation units • No memory overhead for adding optimizations
10. How YJIT works
11. How YJIT works putobject 1 getconst TWO send + leave Bytecode
12. How YJIT works putobject 1 getconst TWO send + leave Bytecode YJIT block1 Mov Reg(0), 1
13. How YJIT works putobject 1 getconst TWO send + leave Bytecode YJIT block1 Mov Reg(0), 1 Context Reg(0): Integer YJIT block2 Mov Reg(1), 2 PatchPoint Constant TWO
14. How YJIT works putobject 1 getconst TWO send + leave Bytecode YJIT block1 Mov Reg(0), 1 Context Reg(0): Integer YJIT block2 YJIT block3 Mov Reg(1), 2 PatchPoint Constant TWO Context Reg(0): Integer Reg(1): Integer PatchPoint Integer#+ Add Reg(0), Reg(1) Ret Reg(0)
15. How ZJIT works putobject 1 getconst TWO send + leave Bytecode
16. How ZJIT works putobject 1 getconst TWO zjit_send + leave Bytecode
17. How ZJIT works putobject 1 getconst TWO zjit_send + leave Bytecode ZJIT HIR v1 = 1 PatchPoint TWO v2 = 2 PatchPoint Integer#+ v3 = 3 Return v3
18. How ZJIT works putobject 1 getconst TWO zjit_send + leave Bytecode ZJIT HIR v1 = 1 PatchPoint TWO v2 = 2 PatchPoint Integer#+ v3 = 3 Return v3 ZJIT LIR Ret 3 PatchPoint Const TWO PatchPoint Integer#+
19. ZJIT IR
20. ZJIT IR • ZJIT IR (Intermediate Representation): • HIR: High-level IR, new in ZJIT • LIR: Ligh-level IR, same as YJIT
21. HIR
22. HIR Optimization Passes
23. LIR
24. LIR Lowering Passes
25. Playing with ZJIT IR • Build: con fi gure --enable-zjit • HIR: ruby --zjit-dump-hir • LIR: ruby --zjit-dump-lir
26. https://tryzjit. fl y.dev/
27. How ZJIT compiles Ruby code
28. How ZJIT compiles Ruby code one: putobject 1 leave two: putobject 2 leave three: putself send :one putself send :two send :+ leave Parse & Compile
29. How ZJIT compiles Ruby code one: putobject 1 leave two: putobject 2 leave three: putself send :one putself send :two send :+ leave one: putobject 1 leave two: putobject 2 leave three: putself zjit_send :one putself zjit_send :two zjit_send :+ leave Parse & Compile Pro fi le
30. How ZJIT compiles Ruby code Initial HIR: fn three@/Users/k0kubun/tmp/a.rb:11: bb0(): EntryPoint interpreter v1:BasicObject = LoadSelf Jump bb2(v1) bb1(v4:BasicObject): EntryPoint JIT(0) Jump bb2(v4) bb2(v6:BasicObject): v11:BasicObject = SendWithoutBlock v6, :one v14:BasicObject = SendWithoutBlock v6, :two v17:BasicObject = SendWithoutBlock v11, :+, v14 CheckInterrupts Return v17 one: putobject 1 leave two: putobject 2 leave three: putself send :one putself send :two send :+ leave Compile
31. How ZJIT compiles Ruby code Optimized HIR: fn three@/Users/k0kubun/tmp/a.rb:11: bb0(): … bb2(v6:BasicObject): PatchPoint MethodRede fi ned(one) PatchPoint NoSingletonClass(Object) v24:HeapObject[Object] = GuardType v6, HeapObject[Object] v31:Fixnum[1] = Const Value(1) PatchPoint MethodRede fi ned(two) PatchPoint NoSingletonClass(Object) v28:HeapObject[Object] = GuardType v6, HeapObject[Object] v33:Fixnum[2] = Const Value(2) PatchPoint MethodRede fi ned(Integer, +) v38:Fixnum[3] = Const Value(3) CheckInterrupts Return v38 one: putobject 1 leave two: putobject 2 leave three: putself send :one putself send :two send :+ leave Inline
32. How ZJIT compiles Ruby code Optimized HIR: fn three@/Users/k0kubun/tmp/a.rb:11: bb0(): … bb2(v6:BasicObject): PatchPoint MethodRede fi ned(one) PatchPoint NoSingletonClass(Object) v24:HeapObject[Object] = GuardType v6, HeapObject[Object] v31:Fixnum[1] = Const Value(1) PatchPoint MethodRede fi ned(two) PatchPoint NoSingletonClass(Object) v28:HeapObject[Object] = GuardType v6, HeapObject[Object] v33:Fixnum[2] = Const Value(2) PatchPoint MethodRede fi ned(Integer, +) v38:Fixnum[3] = Const Value(3) CheckInterrupts Return v38 one: putobject 1 leave two: putobject 2 leave three: putself send :one putself send :two send :+ leave
33. Future of Ruby Performance
34. Future of Ruby Performance • We want less C extensions • In particular, less C → Ruby callbacks
35. C functions dominate execution time Only 10% of execution time is spent in JIT code https://gist.github.com/k0kubun/5e0b3bb894e9fed9b01e25fd25e8bea5
36. C functions dominate execution time Some interpreter implementations are used for reasons https://gist.github.com/k0kubun/5e0b3bb894e9fed9b01e25fd25e8bea5 Complicated argument setup, C → Ruby calls, megamorphic callsite Instance variables: megamorphic callsite
37. C functions dominate execution time 15% is spent on DB queries https://gist.github.com/k0kubun/5e0b3bb894e9fed9b01e25fd25e8bea5
38. C functions dominate execution time 15% is spent on allocation and garbage collection https://gist.github.com/k0kubun/5e0b3bb894e9fed9b01e25fd25e8bea5
39. C functions dominate execution time Many methods are implemented in C https://gist.github.com/k0kubun/5e0b3bb894e9fed9b01e25fd25e8bea5
40. C functions dominate execution time • Methods written in C • C → Ruby calls • Megamorphic callsites Reasons why that happens
41. https://github.com/ruby/ruby/pull/3281
42. Conclusion • We're building ZJIT to unblock cross-instruction optimizations • We want more code to be written in Ruby for performance