Revert r174506

Revert r174506

r174506 replaced toV8Context() with v8::Object::CreationContext(),
and it caused 16% regression in dromaeo.jslibeventjquery.
This is because v8::Object::CreationContext() is slow.

This CL reverts the change.

BUG=383926
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=179085

[haraken, 2014-07-28 22:53:24]

PTAL

[adamk, 2014-07-28 23:14:42]

Have you confirmed this locally? I'm pretty surprised that CreationContext() is
slow. Here's the source of the v8 code involved:

Context* JSObject::GetCreationContext() {
  Object* constructor = this->map()->constructor();
  JSFunction* function;
  if (!constructor->IsJSFunction()) {
    // Functions have null as a constructor,
    // but any JSFunction knows its context immediately.
    function = JSFunction::cast(this);
  } else {
    function = JSFunction::cast(constructor);
  }

  return function->context()->native_context();
}

This should be constant-time, and involve just a few lookups.

[haraken, 2014-07-28 23:51:04]

On 2014/07/28 23:14:42, adamk wrote:
> Have you confirmed this locally? I'm pretty surprised that CreationContext()
is
> slow. Here's the source of the v8 code involved:

Yeah, I can reproduce it. In Linux Release:

toV8Context: 0.31 runs/sec
CreationContext: 0.28 runs/sec

it's a 10% regression. The reported 16% regression in bug 383926 is of Android.

It's indeed true that I optimized toV8Context, but I'm surprised that the
CreationContext code is slower than toV8Context.

[adamk, 2014-07-28 23:55:58]

lgtm, I suppose, but it'd be nice to understand what dromaeo.jslibeventjquery is
doing that makes the slight difference in performance so noticeable. I'd expect
10% slower code to be only a small percent slower in some web-exposed benchmark,
not _even slower_.

[haraken, 2014-07-29 00:11:45]

On 2014/07/28 23:55:58, adamk wrote:
> lgtm, I suppose, but it'd be nice to understand what dromaeo.jslibeventjquery
is
> doing that makes the slight difference in performance so noticeable. I'd
expect
> 10% slower code to be only a small percent slower in some web-exposed
benchmark,
> not _even slower_.

At a first glance at jslib-event-jquery.html, it looks like it's accessing
window.event a lot of times. I'll investigate it more once I finish other tasks
I need to finish.

Let me land this CL to fix the performance regression.

[haraken, 2014-07-29 00:11:48]

The CQ bit was checked by haraken@chromium.org

[commit-bot: I haz the power, 2014-07-29 00:12:27]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/haraken@chromium.org/429433002/20001

[commit-bot: I haz the power, 2014-07-29 02:22:50]

Change committed as 179085

[abarth-chromium, 2014-07-29 17:10:47]

On Mon Jul 28 2014 at 4:14:43 PM <adamk@chromium.org> wrote:

> Have you confirmed this locally? I'm pretty surprised that
> CreationContext() is slow.


We've seen it be slow in the past.  At one point I added a bunch of
machinery to avoid calling it extra times and made dromeo faster.

Adam

To unsubscribe from this group and stop receiving emails from it, send an email
to blink-reviews+unsubscribe@chromium.org.