[webkit-qt] QtWebKit and SunSpider Benchmark: - are these expected results

[Stan Pensak]

Thank you, Jocelyn.  I really appreciate the background information.  It seems unfortunate that it's not so easy to get these details from elsewhere, whether I look at webkit.org, qt-project.org or digia.com.

I just have one more quick (no pun intended) question.  Digia's intro page for Qt 5 highlights the fact that Qt WebKit2 is part of Qt 5.  Is that something I have any way to take advantage of in the Qt widget-based world, or is it only part of Qt Quick and QML?

We are enthusiastic users of Qt with many man years of effort around Qt 3, 4 and now 5.  We want and expect the Qt widget world to maintain feature parity with the QML and Qt Quick approaches.  If this is not going to happen it's probably better that we know sooner latter than later.

Thanks again.
Stan


-----Original Message-----
From: webkit-qt-bounces at lists.webkit.org [mailto:webkit-qt-bounces at lists.webkit.org] On Behalf Of Jocelyn Turcotte
Sent: Tuesday, February 19, 2013 12:33 PM
To: Osztrogonác Csaba
Cc: webkit-qt at lists.webkit.org
Subject: Re: [webkit-qt] QtWebKit and SunSpider Benchmark: - are these expected results

Hi Ossy,

those are cool numbers! I looked at what it would take to get DFG JIT and it seemed like it would take some work.
My main issue is that it would make sense to enable it for 32bit at the same time, and that we should put some testing effort before we release something like this, but we don't have this at reach.

I didn't know that LLINT would give a big improvement in addition to JIT, it's not enabled at all on Windows and it seems like it might be worth looking at before DFG depending on the effort it needs and the risk it introduces.

Jocelyn


On Tue, Feb 19, 2013 at 06:06:45PM +0100, Osztrogonác Csaba wrote:
> Hi,
> 
> Jocelyn, have you tried to enable DFG JIT and/or LLINT on Windows? 
> Both of them can speedup JSC.
> But I have no idea how MSVC friendly the LLINT asm is ...
> 
> On my Linux box I got the following numbers a week before for SunSpider:
> - interpreter (LLINT C-loop):   836 ms
> - LLINT (asm):                  480 ms
> - JIT:                          305 ms
> - LLINT + JIT:                  193 ms
> - LLINT + JIT + DFG-JIT:        140 ms
> 
> Ossy
> 
> Jocelyn Turcotte írta:
> >Hi Stan,
> >
> >Chrome uses V8, while other ports of WebKit, including Qt, use JavaScriptCore (Qt only uses V8 for QML).
> >JavaScriptCore has similar performance to V8 on Mac after Apple implemented what they call DFG JIT.
> >On Windows 32bit it still has basic JIT.
> >On Windows 64bit it lags way behind, where it's still using the interpreter.
> >
> >This patch brings basic JIT support for Windows x64 and should be 
> >released with Qt 5.1: https://bugs.webkit.org/show_bug.cgi?id=107965
> >This should be 2x-4x times faster than the interpreter according to my quick tests.
> >But this still doesn't add support for DFG JIT, still only available on Mac and Linux, which would bring JS performances closer to Chrome's.
> >
> >So yes sadly I think that this kind of performance is currently expected on Windows x64 with Qt 5.0.
> >
> >Cheers,
> >Jocelyn
> _______________________________________________
> webkit-qt mailing list
> webkit-qt at lists.webkit.org
> https://lists.webkit.org/mailman/listinfo/webkit-qt
_______________________________________________
webkit-qt mailing list
webkit-qt at lists.webkit.org
https://lists.webkit.org/mailman/listinfo/webkit-qt