538440 - We never jit DOM workers

Status: VERIFIED FIXED

(Core :: JavaScript Engine, defect)

Description

[Boris Zbarsky [:bzbarsky]]

Bug 533481 made it so workers use separate debugger hooks so we don't call into jsd.  Bug 519719 made it so that we disable the jit on contexts that don't have the runtime's debug hooks.  Result: no jit in workers.

Comment 1

[Mike Beltzner [:beltzner, not reading bugmail]]

Not a release blocker, but we'll relnote it for sure and take it if we do a RC2 or otherwise in 3.6.1, so let's get the fix together.

Comment 2

[Jason Orendorff [:jorendorff]]

Attachment: v1

Comment 3

[Boris Zbarsky [:bzbarsky]]

That looks like it should do the trick, yes.  Want testing or reviews or anything?

Comment 4

[Jason Orendorff [:jorendorff]]

Yes, a test is needed for this.

Comment 5

[Jason Orendorff [:jorendorff]]

Comment on attachment 420832 [details] [diff] [review]
v1

Nobody can think of a good way to test this. Let's get it in, I guess. :-P

Comment 6

[Blake Kaplan (:mrbkap) (inactive)]

Comment on attachment 420832 [details] [diff] [review]
v1

Thanks.

Comment 7

[Jeff Walden [:Waldo]]

Add a js_OnTrace friend API, add a method to DOMWindowUtils that gets called in a loop to check that JITting happens?  Hackish, but adequate.

Comment 8

[Olli Pettay [:smaug][[e-mail]]]

So this didn't make to rc2 after all?

Comment 9

[:Gavin Sharp [email: [e-mail]]]

That's right.

Comment 10

[Jason Orendorff [:jorendorff]]

http://hg.mozilla.org/tracemonkey/rev/502c0b415e10

Comment 11

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/502c0b415e10

Comment 12

[Mike Beltzner [:beltzner, not reading bugmail]]

Comment on attachment 420832 [details] [diff] [review]
v1

a1922=beltzner

Comment 13

[Ben Turner (not reading bugmail, use the needinfo flag!)]

http://hg.mozilla.org/releases/mozilla-1.9.2/rev/858dc58eceee

Comment 14

[Al Billings [:abillings - ex-MoCo]]

Is there an easy way for QA to verify this for 1.9.2 and Trunk?

Comment 15

[Boris Zbarsky [:bzbarsky]]

You could run some script with loops in a worker and see how jitstats are affected (or not) in a debug build....  Or just measure the performance of that script with jit on and off and see whether there's an effect.

Comment 16

[DO NOT USE]

my addon that uses worker threads runs over 2x faster in 3.6.2 than 3.6 (non-debug), so I can only assume they're being jited.

Comment 17

[Mike Shaver (:shaver emeritus)]

Verified! :)

Comment 18

[Al Billings [:abillings - ex-MoCo]]

I agree at this point.

Comment 19

[Heather Arthur [:harth]]

so I guess there were other things that could have sped up workers in 3.6.2, but yeah, trying with jit on/off on the same build has the same speedup.

Comment 20

[Ben Turner (not reading bugmail, use the needinfo flag!)]

Hi harthur,

Just out of curiosity, which addon is it?

Comment 21

[Heather Arthur [:harth]]

https://addons.mozilla.org/en-US/firefox/addon/14328, the thing that uses web workers is the 'website analyzer'.

Comment 22

[Zafar Iqbal]

Hi,

The issue still exist in the current release 3.6.10. Javascript engine can't capture the user change efficiently. 

Only version 3.5.9 firefox is running fine but all above versions have the same issue.

Zafar

Comment 23

[Zafar Iqbal]

Here is the case:

If user remove the character or characters from text field and then enter character and save them (repeat quickly several time), it won't capture the last change.

Zafar

Comment 24

[Boris Zbarsky [:bzbarsky]]

Zafar, that sounds completely unrelated to this bug.  Please file a separate bug on your problem.