Closed Bug 1016908 Opened 10 years ago Closed 10 years ago

10% WinXP CART regression on May 27th on Inbound (fx32) from revision cf90b93d9617

Categories

(Core :: SVG, defect)

x86
Windows XP
defect
Not set
normal

Tracking

()

RESOLVED WONTFIX

People

(Reporter: jmaher, Unassigned)

References

Details

(Keywords: perf, regression, Whiteboard: [talos_regression])

We also got a windows 8 win for CART from the same push.

The patch lowers the accuracy with which we represent invalid regions, so it's a tradeoff between region processing time and painting time. It appears that on win XP (where painting is expensive) this tradeoff is a net loss for CART, whereas on win8 (where painting is cheap) it's a net win.

The overall change depends not only on the speed of painting, but also on the specific page being tested. This change was a *huge* win for the page in the original bug.

We could probably try tune the settings to fix this, but we'd probably lose the win8 win in the process. We also have no idea how we're affecting other pages without a much more comprehensive test suite.

I don't think it's worth investing much time here.
(In reply to Matt Woodrow (:mattwoodrow) from comment #1)
? ... This change was a *huge* win for the page in
> the original bug.
> 
> We could probably try tune the settings to fix this, but we'd probably lose
> the win8 win in the process. We also have no idea how we're affecting other
> pages without a much more comprehensive test suite.
> 
> I don't think it's worth investing much time here.

Agreed. XP is the past and win8 is the future. Unless we're having huge regressions on XP (which we don't here), I think the tradeoff is a good one.
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → WONTFIX
we resolved this as wontfix, but the pgo version saw a 30% regression for winxp.  I am only revisiting this because we see this on Aurora.
You need to log in before you can comment on or make changes to this bug.