Closed Bug 722270 Opened 13 years ago Closed 13 years ago

Nightly (12.0a1) much slower than Beta (10.0) on most V8 benchmark subtests

Categories

(Core :: JavaScript Engine, defect)

12 Branch
ARM
Android
defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla12
Tracking Status
firefox11 --- unaffected
firefox12 + fixed
firefox13 --- unaffected
fennec 12+ ---

People

(Reporter: blechri, Unassigned)

Details

(Keywords: perf)

Latest Nightly 12.0a1 (2012-01-29) is slower than Beta (10.0) on the V8 benchmark (version 6) on my Samsung Galaxy S II. Only subtests RegExp and Splay are faster in Nightly. If I remember correctly Nightly were faster on v8 bench a couple of weeks ago, about as fast as beta is now, so I think there have been a regression but I don't have the numbers I'm afraid. Numbers for Beta Score:769 Richards: 1275 DeltaBlue: 1227 Crypto: 1679 RayTrace: 478 EarleyBoyer: 1057 RegExp: 194 Splay: 616 ------------------------- Numbers for Nightly: Score: 503 Richards: 557 DeltaBlue: 361 Crypto: 669 RayTrace: 396 EarleyBoyer: 618 RegExp: 225 Splay: 1102
Status: UNCONFIRMED → NEW
Ever confirmed: true
tracking-fennec: --- → ?
Keywords: perf
Let's compare XUL and Native 12.0a1 and report back
Assignee: nobody → kbrosnan
This looks to be a Nightly (12a1) only regression atm. Using a Google Nexus Native 12a1 V8 numbers between 465 - 475 XUL 12a1 V8 numbers between 445 - 455 Native 11a2 (Aurora atm, soon Beta) V8 numbers between 835 - 750
Assignee: kbrosnan → nobody
I tested this on the Nightly 12.0a1 builds and found that there is a big difference between the build from 18/01 and the one from 19/01 Here are the numbers: Nightly 12.0a1 (2012-01-18) http://hg.mozilla.org/mozilla-central/rev/f4049f65efc6 Score: 886 Richards: 1298 DeltaBlue: 1196 Crypto: 1678 RayTrace: 436 EarleyBoyer: 1166 RegExp: 184 Splay: 1755 Nightly 12.0a1 (2012-01-19) http://hg.mozilla.org/mozilla-central/rev/58e933465c36 Score: 502 Richards: 543 DeltaBlue: 361 Crypto: 671 RayTrace: 399 EarleyBoyer: 605 RegExp: 235 Splay: 1083 Possible range: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=f4049f65efc6&tochange=58e933465c36
Could this be a regression from bug 718022 or bug 718134?
Sunspider also suffered a regression on Jan 18th, on mozilla-inbound: http://graphs.mozilla.org/graph.html#tests=[[26,63,23],[26,63,20]]&sel=none&displayrange=30&datatype=running The regression was caused by bug 696291: http://hg.mozilla.org/integration/mozilla-inbound/rev/095649e65552 That regression seems to have been fixed by bug 696291 (different patch): https://hg.mozilla.org/integration/mozilla-inbound/rev/0aacfba4caab We don't run V8 on mobile talos, but you might be able to confirm using a tinderbox build from mozilla-inbound, if it still exists: http://stage.mozilla.org/pub/mozilla.org/mobile/tinderbox-builds/mozilla-inbound-android/1326855029/
(In reply to Mark Finkle (:mfinkle) from comment #5) > We don't run V8 on mobile talos, but you might be able to confirm using a > tinderbox build from mozilla-inbound, if it still exists: > http://stage.mozilla.org/pub/mozilla.org/mobile/tinderbox-builds/mozilla- > inbound-android/1326855029/ Confirm that this build regressed V8 I am CC'ing some JS devs to see if they have thoughts on why V8 was affected, and why it was not "fixed" by the latest patch landing from bug 696291.
Assignee: nobody → general
tracking-fennec: ? → 11+
Component: General → JavaScript Engine
Product: Fennec Native → Core
QA Contact: general → general
tracking-fennec: 11+ → 12+
Just tried again with latest Nightly and it's fast again. Nice. Score: 836 Richards: 1179 DeltaBlue: 1220 Crypto: 1432 RayTrace: 447 EarleyBoyer: 1087 RegExp: 201 Splay: 1417
Aaron - can you help us find a regression/degression range so that we can uplift the latest fix on FF13 or come up with a low risk fix?
The regression range is in comment 3 and comment 5
Can we close it as WFM based on comment 7?
Status: NEW → RESOLVED
Closed: 13 years ago
Resolution: --- → WORKSFORME
Resolution: WORKSFORME → FIXED
Target Milestone: --- → mozilla12
Version: Trunk → 12 Branch
Removing qawanted as per comment 9
Keywords: qawanted
Whiteboard: [qa+]
Whiteboard: [qa+]
You need to log in before you can comment on or make changes to this bug.