994683 - 2% canvasmark regression on osx 10.6 from inbound revision 56e8252c5467

Status: RESOLVED INVALID

(Core :: Graphics, defect)

Description

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

ironically this is a patch to fix a regression but it seems to have caused a 2% regression in tcanvasmark only on osx 10.6:
http://graphs.mozilla.org/graph.html#tests=[[297,63,21]]&sel=1396927539000,1397100339000&displayrange=7&datatype=running

I did some retriggers and the values pre/post this patch are reliable (no noise causing the patch to be incorrectly identified):
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&fromchange=08aa13c6feac&tochange=ddea7fdd09bf&jobname=talos%20chromez

Comment 1

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

in bug 990490 we discuss how much value there is to work on 10.6 only regressions (especially a small one).  The exception is graphics work, I am not sure if that is something to be concerned with in this regression or not.

Comment 2

[Benjamin Kerensa [:bkerensa]]

Where did this regress exactly? Also can you explain impact to our users if we chose not to track or block on this.

Comment 3

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

canvasmark is a 2d canvas benchmark, a 2% regression is small and on osx 10.6 only limits this.  I am asking Jeff for info as he might know more about what happens in bug 990854 to cause this.

Comment 4

[Jeff Muizelaar [:jrmuizel]]

I don't see how this could have caused a regression. It should have improved things.

Comment 5

[Jeff Muizelaar [:jrmuizel]]

Attachment: An example benchmark of the jist of the patch

I tested this on 10.6 and the new version was faster

Comment 6

[Jeff Muizelaar [:jrmuizel]]

I'm not sure that there's much that can be done here.

Comment 7

[Avi Halachmi (:avih)]

In canvasmark, higher values are better: https://wiki.mozilla.org/Buildbot/Talos/Tests#CanvasMark and the value went up.

Sorry for the incorrect regression report.

Comment 8

[Sylvestre Ledru [:Sylvestre]]

Invalid, untracking.