[ Firefox 3.6.4 ] Poor performance in path tracing example (using WebWorkers)

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[ Firefox 3.6.4 ] Poor performance in path tracing example (using WebWorkers)

Kr@n

Hi,

I recently came across an implementation of path tracing in
javascript, using web workers. Since the developer himself said
performance was noticeably better in Chrome, I tried FF3.6.4 beta and
Chrome 4.1.249.1064 (45376) and Chrome is at least 10 times faster (on
an Intel E8500 CPU, 2GB RAM, but I don't think this is relevant) ...

It may be caused by poor tracing in FF (JaegerMonkey would thus kick
in and close the gap), but in case it's not, here is the URL :
http://29a.ch/sandbox/2010/cornellbox/

Hope that helps,


Matthieu.
_______________________________________________
dev-performance mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-performance
Reply | Threaded
Open this post in threaded view
|

Re: [ Firefox 3.6.4 ] Poor performance in path tracing example (using WebWorkers)

Dave Mandelin-2
On 5/25/2010 4:25 AM, Kr@n wrote:

>
> Hi,
>
> I recently came across an implementation of path tracing in
> javascript, using web workers. Since the developer himself said
> performance was noticeably better in Chrome, I tried FF3.6.4 beta and
> Chrome 4.1.249.1064 (45376) and Chrome is at least 10 times faster (on
> an Intel E8500 CPU, 2GB RAM, but I don't think this is relevant) ...
>
> It may be caused by poor tracing in FF (JaegerMonkey would thus kick
> in and close the gap), but in case it's not, here is the URL :
> http://29a.ch/sandbox/2010/cornellbox/

I think it's not tracing:

Recording starting from
http://29a.ch/sandbox/2010/cornellbox/worker.js:194@100
(FragID=000000)
Abort recording of tree
http://29a.ch/sandbox/2010/cornellbox/worker.js:194@100
at http://29a.ch/sandbox/2010/cornellbox/worker.js:201@0: Inner tree is
an unsupported type of recursion.

Could you try recoding it to use iteration instead of recursion in
Renderer.trace and see what that does to perf?

Dave
_______________________________________________
dev-performance mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-performance
Reply | Threaded
Open this post in threaded view
|

Re: Poor performance in path tracing example (using WebWorkers)

Kr@n
On 25 mai, 23:19, David Mandelin <[hidden email]> wrote:

> I think it's not tracing:
>
> Recording starting fromhttp://29a.ch/sandbox/2010/cornellbox/worker.js:194@100
> (FragID=000000)
> Abort recording of treehttp://29a.ch/sandbox/2010/cornellbox/worker.js:194@100
> athttp://29a.ch/sandbox/2010/cornellbox/worker.js:201@0:Inner tree is
> an unsupported type of recursion.

Thanks for your prompt answer. I was pretty sure myself this was due
to tracing abortion (path tracing algorithms are implemented
recursively more often than not), but I wanted to be sure ...


> Could you try recoding it to use iteration instead of recursion in
> Renderer.trace and see what that does to perf?

Unfortunately, I am not the developer of this path tracing test (I
just happen to shudder every time I read someone claiming Chrome is
immensely faster than FF).
However I will make sure to tell the author of this example to
implement his algo using iteration if he wants solid performance on FF
before JaegerMonkey makes it to an official release ...

Thanks again for your time.

Matthieu.
_______________________________________________
dev-performance mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-performance
Reply | Threaded
Open this post in threaded view
|

Re: Poor performance in path tracing example (using WebWorkers)

Dave Mandelin-2
On 5/26/2010 6:27 AM, Kr@n wrote:
> On 25 mai, 23:19, David Mandelin<[hidden email]>  wrote:
>
> Unfortunately, I am not the developer of this path tracing test (I
> just happen to shudder every time I read someone claiming Chrome is
> immensely faster than FF).
> However I will make sure to tell the author of this example to
> implement his algo using iteration if he wants solid performance on FF
> before JaegerMonkey makes it to an official release ...

Thank you. By the way, I'd love to have a version of this site that we
can use as a benchmark: something that shows perf numbers, perhaps
frames per second, or some rendering time metric. Could you put me in
touch with the author, or ask for a version like that?

Dave
_______________________________________________
dev-performance mailing list
[hidden email]
https://lists.mozilla.org/listinfo/dev-performance