JSX Optimizer
9 likes6,771 views
The document discusses the JSX Optimizer, which performs optimizations on JavaScript code that JavaScript VMs cannot. It aims to minimize the need for inline caching, pressure on the garbage collector, and maintain the original code structure while optimizing. Some optimizations included are constant folding, inlining functions, and unboxing. Challenges include switching to SSA form and maintaining debuggability of the original code. Benchmark results show a 13.5-28.7% increase in frames per second for the Box2D game engine.
![Inline Caching
The idea:
most of the code will be called with the same
types of arguments
cache the expectation at the call site
Example: point.x
type of point will be { x: ?, y: ? }
thus if the expectation succeeded, point.x will be:
load point_addr[offset_of_x]
Jul 19 2012 JSX Optimizer 5](https://image.slidesharecdn.com/jsx-optimization-120719104046-phpapp01/85/JSX-Optimizer-5-320.jpg)
JSX Optimizer
- 1. JSX Optimizer DeNA Co., Ltd. Kazuho Oku
- 2. The Basic Idea of the JSX Optimizer perform optimizations that JavaScript VMs cannot Jul 19 2012 JSX Optimizer 2
- 3. Benchmark Scores Box2D.jsx +13.5% (iOS 5.0, iPhone 4) +11.8% (iOS 5.1, iPhone 4S) +28.7% (Android 2.3, Xperia ray) numbers are increase in fps ToDo: post the details Jul 19 2012 JSX Optimizer 3
- 4. The Nature of JavaScript a dynamic language primitives are untyped a + 1 becomes "a1" if a is "a", 2 if a is 1 objects are untyped a.p may point to whatever solution: inline caching V8, JavaScriptCore, SpiderMonkey Jul 19 2012 JSX Optimizer 4
- 5. Inline Caching The idea: most of the code will be called with the same types of arguments cache the expectation at the call site Example: point.x type of point will be { x: ?, y: ? } thus if the expectation succeeded, point.x will be: load point_addr[offset_of_x] Jul 19 2012 JSX Optimizer 5
- 6. Guards in Inline Caching Example: a = b.p + 1 typeof(b) is as expected load b.p typeof(b.p) is number add one store the result Jul 19 2012 JSX Optimizer 6
- 7. Guards in Inline Caching (cont'd) Example: f() typeof(f) is function setup stack frame jump to the code of f Jul 19 2012 JSX Optimizer 7
- 8. Guards in Inline Caching (cont'd) Example: f() // after inlined typeof(f) is function f has not been altered execute the code expanded in-line Jul 19 2012 JSX Optimizer 8
- 9. Guards in Inline Caching (cont'd) Example: o.f() typeof(o) is as expected load o.prototype typeof (o.prototype) is as expected load o.prototype.f typeof(f) is function setup stack frame jump to the code of f Jul 19 2012 JSX Optimizer 9
- 10. Guards in Inline Caching (cont'd) Example: o.f() // after inlined typeof(o) is as expected load o.prototype typeof (o.prototype) is as expected load o.prototype.f typeof(f) is function f has not been altered execute the code expanded in-line Jul 19 2012 JSX Optimizer 10
- 11. What interpreters do / do not inter-statement optimization V8 does SpiderMonkey and JavaScriptCore seem not inline-expand the JavaScript functions at least V8 does Jul 19 2012 JSX Optimizer 11
- 12. JSX optimizer minimizes the need for IC caches property access JSX does not have accessors (Object.defineProperty) inline-expands the functions converts member methods to static functions generates IC-friendly code always initializes the properties in same order Note: { x: 0, y: 0 } and { y: 1, x: 1 } are different types Jul 19 2012 JSX Optimizer 12
- 13. JSX optimizer minimizes the pressure on GC unboxing folds constants JavaScript does not have a true const Jul 19 2012 JSX Optimizer 13
- 14. Example: affine transformation input: new Matrix(1, 0, 0, 0, 2, 0).transform(new Point(x, y)) output: {x: x + 0 * y, y: 0 * x + 2 * y} Note: 0 * n cannot be optimized to 0, since it becomes NaN if n is NaN Jul 19 2012 JSX Optimizer 14
- 15. Current Optimization Subcommands lto no-assert no-log unclassify fold-const inline dce return-if lcse unbox array-length Jul 19 2012 JSX Optimizer 15
- 16. Diffs. from an ordinary optimizing compiler ordinary compiler generate a low-level non-redundant code JSX compiler most code are left as is while optimizing the code wherever possible Jul 19 2012 JSX Optimizer 16
- 17. The Difficulties and Considerations switch to SSA-based code? might end-up in generating a very different code from source would be difficult to debug (by the users) may become slow (than the same-looking code in JavaScript); GCC has the problem it is troublesome to optimize the code while maintaining the original code structure example. support various kind of loops: do-while, for, for..in, while Jul 19 2012 JSX Optimizer 17
- 18. The Difficulties and Considerations (cont'd) Other optimization techniques? inline allocation? more powerful dce / dse reduce the number of locals? Jul 19 2012 JSX Optimizer 18
- 19. Some Microbenchmarks http://jsperf.com/local-vs-member/2 http://jsperf.com/construct-object-vs-array/3 http://jsperf.com/inherit-object-vs-array-set http://jsperf.com/vars-in-nested-closure Jul 19 2012 JSX Optimizer 19