482853 - Garbage collection correlates with pauses in execution

Status: RESOLVED INVALID

(Core :: JavaScript: GC, defect)

Description

[James Boston (:jboston) <[e-mail]>]

User-Agent:       Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-GB; rv:1.9.0.7) Gecko/2009021906 Firefox/3.0.7
Build Identifier: Mozilla/5.0 (Macintosh; U; Intel Mac OS X 10.5; en-GB; rv:1.9.0.7) Gecko/2009021906 Firefox/3.0.7

Small pauses in execution caused by garbage collection are visible in applications that try to render smooth animations on a canvas. (I've tried putting some printf statements into Mozilla's garbage collection code to verify this and I'm certain that gc is the cause of the pause.) Our app updates every 25ms using setInterval, but there are semi-frequent pauses where intervals are 250ms or longer.

We'd like to minimize the creation of garbage by re-using arrays of doubles, but from what I understand, changing the value of a double causes the allocation of new memory.

It would be helpful if the engine could minimize the creation of garbage when using doubles.

Reproducible: Always

Steps to Reproduce:
1. In about:config set browser.dom.window.dump.enabled = true 
2. Restart browser from command line
3. If you don't have the canvas3d extension installed select "Use dummy canvas context"
4. Select "Dump to console"
5. Run the benchmark with a time of 60 seconds or greater.
Actual Results:  
One frame will sometimes take much longer than the rest to render:
30 31 32 39 30 36 32 35 396 20 
29 37 33 34 24 25 27 29 30 32 
26 
---> FPS: 21



Visit c3dl.org for more information on how we are creating a 3D javascript library for canvas3d.

Comment 1

[James Boston (:jboston) <[e-mail]>]

If it's not obvious the URL for the benchmark:
http://www.c3dl.org/test-case/benchmark.html

Comment 2

[James Boston (:jboston) <[e-mail]>]

Similar bug reports:
https://bugzilla.mozilla.org/show_bug.cgi?id=482204
https://bugzilla.mozilla.org/show_bug.cgi?id=408372#c4

Comment 3

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

I'd suggest creating a html page with just a button that when pushed runs the relevant test (without canvas3d), and displays the results in the page... or even better yet, something that can run in the standalone shell.

Comment 4

[James Boston (:jboston) <[e-mail]>]

Attachment: Stand alone test

I've nicked the code from bug 482204's test and modified it slightly to try to simulate the pushing and pulling from the matrix stack that occurs in the c3dl library. I've also changed it so that the array that contains the matrix is scoped in a way that it could be re-used if that were possible. Hopefully this simulates the action of the c3dl. Regardless, it does capture gc pauses.

Comment 5

[Wayne Mery (:wsmwk)]

can be closed?

Comment 6

[Jon Coppeard (:jonco)]

We have enough other bugs on file for this sort of thing.