Javascript Performance Thread

[sciguyryan]

Of course, these things are written in JS.

Are you sure? I remember activating TM for chrome content was delayed quite a bit when it was new. Is this different for JM?

I think method JIT is intended on being enabled for everything from the offset.

[bzbarsky]

is JM/TM speeding up any part of Tab Candy (Firefox Panorama) or the Extensions (Chrome) themselves yet ?

Tab Candy is mostly graphics-intensive, not JS-intensive. So JM wouldn't help it much anyway, whereas the GPU work will.

At the moment JM is not on for chrome by default yet, but the plan is to flip the switch in the next few days.

[WonderCsabo]

Tab Candy is mostly graphics-intensive, not JS-intensive. So JM wouldn't help it much anyway, whereas the GPU work will.

Interesting. I use GPU acceleration (D2D, layers), and TabCandy animations are still very sluggish. I hoped JM will help...

[sabret00the]

Why is the JM/TM merge happening prior to beating Apple/Google's speed on arewefastyet.com

[WonderCsabo]

They have to merge it to m-c now, so they have enough time to explore and fix potential bugs. I think they will further optimize the engine after the merge, too.

[sciguyryan]

Why is the JM/TM merge happening prior to beating Apple/Google's speed on arewefastyet.com

For testing reasons. There will probably be several more merges before the final version to pick up further improvements but it needs to go on soon to allow for testing.

[Dorus]

Why is the JM/TM merge happening prior to beating Apple/Google's speed on arewefastyet.com

Beside, i doubt they will be delaying 4.0 because the javascript engine isn't as fast as opera/chrome yet. At best they will push JM updates with minor updates, trough i dind't see any concrete wording about that yet other then that they will be shipping non-user experience breaking changes to minor releases more and more.

[Cru_N_cher]

Why is the JM/TM merge happening prior to beating Apple/Google's speed on arewefastyet.com

For testing reasons. There will probably be several more merges before the final version to pick up further improvements but it needs to go on soon to allow for testing.

another thing is they trying to be more sane and don't put every effort into synthetic benchmarks anymore but more into real life performance, currently working very closely with MSR on this topic

[mikeclackler]

FYI JaegerMonkey has landed on the trunk

[WonderCsabo]

I'm not seeing it in the hourlies, yet. (Or am is doing something wrong?)

[_Alexander]

I'm not seeing it in the hourlies, yet. (Or am is doing something wrong?)

You are not patient.

[mikeclackler]

I'm not seeing it in the hourlies, yet. (Or am is doing something wrong?)

hourly builds are not finished, watch for a green "B" for your operating system for the large JM merge:
http://tests.themasta.com/tinderboxpushlog/

[panocmp]

about performance sunspider gives me better results by disabling trarcemonkey for content

pre JM
RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 737.4ms +/- 4.4%
--------------------------------------------

with JM+TM
ESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 611.6ms +/- 4.0%
--------------------------------------------

JM only[
RESULTS (means and 95% confidence intervals)
--------------------------------------------
Total: 569.8ms +/- 3.7%
--------------------------------------------

CPU T5500, Intel X3100, D3D+D2D enabled

[Gusar]

@panocmp: Yep, that's consistent with the arewefastyet.com graphs. It's a matter of tuning the heuristics.

[wellofsouls]

Why is the JM/TM merge happening prior to beating Apple/Google's speed on arewefastyet.com

because it's already a lot faster than TM alone? I don't think they said anything about landing JM+TM only after it's faster than Apple/Google's speed, but that there's no regression from TM.