Bug 103331 – [V8] Replace toWebCoreString()/toWebCoreAtomicString() with V8Parameter

• Status: RESOLVED
• Resolution: INVALID
• Component: WebCore JavaScript

[Kentaro Hara]

V8ParameterBase is slow. V8ParameterBase is being used in hot call paths.

  // Code A
  AtomicString str = toWebCoreAtomicString(v8Object); // Assume that v8Object is v8::String or v8::Int32
  imp->setAttribute(HTMLNames::idAttr, str);

  // Code B
  STRING_TO_V8PARAMETER_EXCEPTION_BLOCK(V8Parameter<>, str, v8Object); // Assume that v8Object is v8::String or v8::Int32
  imp->setAttribute(HTMLNames::idAttr, str);

Code B is 10% slower than Code A. Ideally Code A should be as fast as Code B.

Then we can replace all toWebCoreAtomicString()s and toWebCoreString()s with V8ParameterBase; i.e. we can always use V8ParameterBase regardless of String or AtomicString.

[Kentaro Hara]

Created attachment 176130 [details]
micro benchmarks

Hmm, it looks like 10% slowness was placebo. The perf is a bit flaky.

As far as I test with the micro benchmarks attached to this bug and Bindings/id-setter.html a lot of times, there is no observable perf diff between V8Parameter and toWebCoreString()/toWebCoreAtomicString().

So the conclusion is that V8Parameter is as fast as toWebCoreString()/toWebCoreAtomicString(), and thus we can replace toWebCoreString()/toWebCoreAtomicString() with V8Parameter without regressing performance.

[Brian Burg]

Closing some V8-related work items.