Closed Bug 1201470 Opened 9 years ago Closed 9 years ago

Regression on octane e-b/deltablue/splay

Categories

(Core :: JavaScript Engine: JIT, defect)

defect
Not set
normal

Tracking

()

RESOLVED FIXED

People

(Reporter: h4writer, Unassigned)

References

Details

A regression/improvement was reported on AWFY:
- slave: Mac OS X 10.10 32-bit (Mac Pro, shell)
- mode: Ion

Regression(s)/Improvement(s):
- octane: DeltaBlue: -5.22% (regression)
- octane: EarleyBoyer: -9.28% (regression)
- octane: Splay: -23.23% (regression)

Recorded range:
- http://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=45f2da1082dd&tochange=71886d218c12

More details: http://arewefastyet.com/regressions/#/regression/1796104

Manual bisection on the e-b benchmark gave me:

The first bad revision is:
changeset:   260540:c9e469c6b915
user:        Terrence Cole <terrence@mozilla.com>
date:        Thu Aug 20 10:35:22 2015 -0700
summary:     Bug 1196847 - Part 1: Allow storage of a unique id for a cell independent of address; r=jonco
Terrence, could this be caused by bug 1196847 ?
Flags: needinfo?(terrence)
Blocks: 1196847
It has to be that patch, although I'm a bit baffled as to how. All it's effectively doing at the moment is adding a could words to the Zone. Maybe it's changed the alignment of something downstream? Still I cannot image how that could result in a 23% slowdown. I'll backout after this mornings meeting and see if I can reproduce locally.
(In reply to Terrence Cole [:terrence] from comment #2)
> It has to be that patch, although I'm a bit baffled as to how. All it's
> effectively doing at the moment is adding a could words to the Zone. Maybe
> it's changed the alignment of something downstream? Still I cannot image how
> that could result in a 23% slowdown. I'll backout after this mornings
> meeting and see if I can reproduce locally.

Thanks for the quick action! Confirmed fixed
Status: NEW → RESOLVED
Closed: 9 years ago
Flags: needinfo?(terrence)
Resolution: --- → FIXED
(In reply to Terrence Cole [:terrence] from comment #2)
> It has to be that patch, although I'm a bit baffled as to how. All it's
> effectively doing at the moment is adding a could words to the Zone. Maybe
> it's changed the alignment of something downstream? Still I cannot image how
> that could result in a 23% slowdown. I'll backout after this mornings
> meeting and see if I can reproduce locally.

This regression was ultimately from the CAS I added to JSObject::finalize. Even without any further instructions, this appears to destroy performance, probably by forcing unnecessary synchronization.
You need to log in before you can comment on or make changes to this bug.