How does V8 optimize JavaScript code?

Transcript

1. How does V8 optimize JavaScript code?
2. Who am I? • Developer Advocate at Onix-Systems, Kirovohrad; • Kittik author (engine for ASCII presentations); • NodeJS outside collaborator; • Contributor in Sails ecosystem (in the past); • Open source maniac;
3. Before we begin, let me start the holy war
4. C++ is ~5 times faster than JavaScript
5. Or is it?
6. JavaScript is 17% slower than optimized C++
8. How is this possible?
9. Let the story begin
10. JS -> AST
11. –Wikipedia “AST is a tree representation of the abstract syntactic structure of source code. Each node of the tree denotes a construct occurring in the source code.”
12. AST -> Full Codegen Full Codegen is a generic compiler in V8. The main purpose of it is to compile JavaScript to native code as quick as possible, generating the code without optimizations. It takes AST, walks over the nodes and emits calls to a macroassembler directly.
13. AST -> Full Codegen
14. It works, runs fast, but doesn’t execute fast
15. We need to get type feedback about arguments\variables
16. Profiling thread alongside with Full-Codegen When running your application, V8 also runs profiling thread. Profiling thread collects information about your functions, arguments and types that go through them.
17. We have enough information about types
18. Imagine that V8 add type declarations into code
19. AST and type feedback goes into Crankshaft
20. Crankshaft compiler Crankshaft compiler is an optimization compiler in V8. It takes AST and type feedback and compiles optimized code for your function. Optimized code of function replaces un-optimized one using on-stack replacement (OSR) then.
21. Crankshaft consists of Hydrogen and Lithium
22. Hydrogen compiler Takes AST and type feedback and generates high- level intermediate representation (HIR) which consists of a control-flow graph (CFG) in static single assignment form (SSA). During the HIR generation, actual optimizations are applied, such as constant folding, method inlining, dynamic type feedback, etc… The result is an optimized CFG that is used as input to the next compiler - Lithium.
23. AST + TF -> Hydrogen
24. Lithium compiler Lithium compiler takes optimized HIR and translates it to machine-specific low-level intermediate representation (LIR). During LIR generation, Crankshaft can still apply some low-level optimizations to it. After LIR was generated, Crankshaft generates a sequence of native instructions for each Lithium instruction and gets executed.
25. HIR -> Lithium
26. LIR -> native instructions
27. JavaScript AST Full Codegen Native Instructions Profiling Thread Hydrogen Lithium
28. Summary • JavaScript is fast enough, comparing to C++; • Full Codegen compiler compiles JavaScript to native code without optimizations; • Profiling thread collects information about types, allowing to optimize your code; • Crankshaft is a black box where optimizations are applied;
29. That’s how :)
30. Thanks! ghaiklor blog.ghaiklor.com ghaiklor