1037516 - 2.1% osx 10.6 kraken regression on inbound (fx33) from push baec82f2baeb on July 8

Status: RESOLVED FIXED

(Core :: JavaScript: GC, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

Here is a graph to show the problem:
http://graphs.mozilla.org/graph.html#tests=[[232,63,21]]&sel=1404743285000,1404916085000

I did some retriggers to help show the failure:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=e5c88681e091&tochange=b94c916122c8&jobname=Rev4%20MacOSX%20Snow%20Leopard%2010.6%20mozilla-inbound%20talos%20dromaeojs

we end up at this changeset:
http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=c6ca2340a718&tochange=baec82f2baeb

The problem seems to be in audio-beat-detection:
http://people.mozilla.org/~klahnakoski/talos/Alert-Results.html#sampleMin=2014-07-07&platform=x86_64&sampleMax=2014-07-10&branch=Fx-Team&test=audio-beat-detection&os=mac.OS+X+10.6.8



this appears to be osx 10.6 only.

Comment 1

[Jon Coppeard (:jonco)]

Removing bug 1034566 as that code isn't normally built.

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

so this is related to bug 988486 or more specifically:
https://hg.mozilla.org/integration/mozilla-inbound/rev/baec82f2baeb

Comment 3

[Jon Coppeard (:jonco)]

Attachment: bug1037516-talos

Bug 988486 is just refactoring, and almost none of the code touched is hot.

The only thing I can think of is that as GCRuntime::updateBytesAllocated() is marked inline only in the class declaration and not in the definition then clang might not be inlining it and that might cause this regression.

Here's a patch to test this.

Comment 4

[Jon Coppeard (:jonco)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/f1a8a13674dd

Comment 5

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/f1a8a13674dd

Comment 6

[Joel Maher ( :jmaher ) (UTC -8)]

thanks, this regression appears to be fixed!

Comment 7

[Jon Coppeard (:jonco)]

Great, let's close this.