Closed Bug 1447391 Opened 7 years ago Closed 7 years ago

Excessive time spent in GCCellPtr::mayBeOwnedByOtherRuntimeSlow()

Categories

(Core :: XPCOM, defect)

defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla61
Tracking Status
firefox-esr60 --- wontfix
firefox60 --- wontfix
firefox61 --- fixed

People

(Reporter: sfink, Assigned: sfink)

References

Details

Attachments

(1 file)

jesup was looking at a profile that takes 863ms out of a several-second CC in the above function.
We only need to do mayBeOwnedByOtherRuntimeSlow for strings and symbols, neither of which can be a member of a cycle.
Attachment #8960700 - Flags: review?(continuation)
Comment on attachment 8960700 [details] [diff] [review] Early return for strings and symbols when tracing for CC Review of attachment 8960700 [details] [diff] [review]: ----------------------------------------------------------------- Thanks for investigating and fixing this.
Attachment #8960700 - Flags: review?(continuation) → review+
Pushed by sfink@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/a65fcbcfe819 Early return for strings and symbols when tracing for CC, r=mccr8
Status: ASSIGNED → RESOLVED
Closed: 7 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla61
[Tracking Requested - why for this release]: Users are seeing severe performance problems in Firefox 60 that would be helped by this patch. It isn't clear if this is a complete solution or not. (for example: https://www.reddit.com/r/firefox/comments/8if1fs/anyone_else_seeing_massive_cpu_spikes_with_ff_60/ )
See Also: → 1460385
One user on Reddit was still experiencing problems with Firefox 62, so this may not fix the worst of it. https://www.reddit.com/r/firefox/comments/8if1fs/anyone_else_seeing_massive_cpu_spikes_with_ff_60/dyrh099/
Per IRC discussion with mccr8, it doesn't appear that we need to backport this to 60 after all.
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: