Perf Comparison (plaintext)

[JoeS]

Well, I finally found a way to do some basic performance comparisons using
sunspider-0.9 Tests

-------------------------------------------
Total: 2.0.0.20pre (20081217) 47212.0ms +/- 1.1%
--------------------------------------------

--------------------------------------------
Total: Shredder/3.0b1pre ID:20080920 14423.8ms +/- 12.0%
--------------------------------------------

--------------------------------------------
Total: Shredder/3.0b2pre ID:20090218 4758.6ms +/- 3.6%
--------------------------------------------

I would say that's real progress.
Test details follow in html

JoeS

[Gary Kwong]

I'm not an expert in this, but I suspect we're picking up JS perf
improvements from the Platform on 1.9.1 (which 3.0b1pre and 3.0b2pre are
based on), vs 2.0.0.20pre which remains on 1.8.1.

That said, once heavyweight patch in bug 452498 lands on 1.9.1 in the
coming weeks, our numbers might look even better.

I wonder how much faster/better TB will feel like, if we turned on JIT
in TB chrome.

-Gary
nth10sd

[David Ascher]

On 2/18/09 9:35 PM, Gary Kwong wrote:
> I wonder how much faster/better TB will feel like, if we turned on JIT
> in TB chrome.

It's an easy test to make. When I did it last time, I didn't notice much.

The expectation from what I understand is that much of the
perf-sensitive areas that involve JS involve XPConnect, so wouldn't be
fast-pathed by the JITter w/o a lot of custom work.

--david

[Philip Chee]

On Thu, 19 Feb 2009 13:35:36 +0800, Gary Kwong wrote:

> I wonder how much faster/better TB will feel like, if we turned on JIT
> in TB chrome.

If you turn on JIT in chrome, watch out for Bug 473552 (TM: SIGILL due
to TraceMonkey emitting unsupported NOPL) if you have some weird CPU
like Via C3, Via Eden, AMD Geode LX (as used in OLPC), Transmeta Crusoe

Phil

--
Philip Chee <phi...@aleytys.pc.my>, <phili...@gmail.com>
http://flashblock.mozdev.org/ http://xsidebar.mozdev.org
Guard us from the she-wolf and the wolf, and guard us from the thief,
oh Night, and so be good for us to pass.
[ ]No, Q, I meant a BUD light!
* TagZilla 0.066.6

[Matthew Monaco]

What platform were you testing this on?

[Robert Kaiser]

Gary Kwong wrote:
> I wonder how much faster/better TB will feel like, if we turned on JIT
> in TB chrome.

From what I heard so far, it would at least be remarkably faster in
crashing ;-)

Robert Kaiser

[Dan Mosedale]

On 2/18/09 8:54 PM, JoeS wrote:
> Well, I finally found a way to do some basic performance comparisons using
> sunspider-0.9 Tests
>

Thanks for generating that data; it's definitely helpful to be able to
see our trajectory there. How much work was it to generate?

Dan

[JoeS]

On 2/19/2009 8:49 AM, Matthew Monaco wrote:
> What platform were you testing this on?

Winxp with a rather slow 1.4 gig Pentium III-s cpu

[JoeS]

Actually, the whole series of tests is automated on the webkit.org site.
Took me longer to format the specific data (in the html version) than to actually run the tests.

And..no.. Javascript does not run within "message" context in current trunk, your patch worked fine.

Joe

[Dan Mosedale]

On 2/19/09 2:04 PM, JoeS wrote:
> On 2/19/2009 1:47 PM, Dan Mosedale wrote:
>> Thanks for generating that data; it's definitely helpful to be able to
>> see our trajectory there. How much work was it to generate?
>

> Actually, the whole series of tests is automated on the webkit.org site.
> Took me longer to format the specific data (in the html version) than to
> actually run the tests.

Nice. Something that would be even more valuable, I think, would be
performance data in and around UI responsiveness (eg loading messages,
changing folders, etc). Do you have any interest in poking around in
that area?

> And..no.. Javascript does not run within "message" context in current
> trunk, your patch worked fine.

Glad to hear that; thanks for the feedback.

Dan