[webkit-dev] relative JS speed between Safari/Firefox

[Mike Emmel]

I was looking at the Linux KDE numbers KJS where it is a lot slower.
I actually did not see the last number till later after the email was sent.
I was looking at the treecode branch.

Its impressive I did not believe it was possible to get a AST to
perform on par with bytecode you learn something every day
historically there 5-10 times slower.

Its seems I was wrong and gave the link to prove it :)


In this case Firefox1.5 on Linux with a score of 24 should be compared
to Safari 2.0 with a 22 and indeed the performance is on par.
Note that  Opera is a 10.  Now with a bytecode interpeter Webkit
should be down with Opera maybe even better.


I do agree now that the latest numbers look good. With that said my
fundamental concern about the AST approach is it does not scale well
to thousands of lines of javascript thats starting to show up in
current AJAX apps but if apples getting these numbers with a AST I'm
intrested to see the results with a optimized bytecode implementation.

Mike

On 5/26/06, John Sullivan <sullivan at apple.com> wrote:
> I think Maciej's earlier message about this topic did an excellent job
> of explaining the different issues here. I just wanted to add that as
> far as I can tell, the howtocreate link below seems to show that
> Safari 2.0 is faster than Firefox in every listed category other than
> "Multiple images", in which Firefox 1.5 is a little faster than Safari
> 2.0. For "Script speed" it gives:
>
> Firefox 1.0 (krmathis)  53
> Firefox 1.0                     72
> Firefox 1.5                     40
> Safari 2.0                      27
>
> where lower numbers are faster.
>
> So I'm not sure why you claim that this page shows Firefox as
> significantly faster on scripts than Safari.
>
> John
>
> On May 26, 2006, at 3:08 PM, Mike Emmel wrote:
>
> > Finally found some benchmarks form what I can tell Firefox is
> > significantly faster on scripts
> > then Safari
> >
> > http://www.howtocreate.co.uk/browserSpeed.html#testresults
> >
> > It does look like the khtml engine itself is quite fast so the
> > combination of SpiderMonkey with
> > WebCore should make a very fast browser.
> >
> > Again please prove that KJS is faster then SpiderMonkey everything I
> > finds shows not only is this not true but it is significantly slower.
> > I see no reason that the proposal to integrate SpiderMonkey slim it
> > down and do a little performance tunning is not valid and resonable
> > approach. I see no reason to start from scratch on a new interpeter.
> >
> > Finally I think this benchmark should be run agianst webkit/IE/Firefox
> > and the results should be on the website it would be helpful. It took
> > me a while to find a page with a reasonable set of results.
> >
> > Mike
>
>