544477 - Regression in Peacekeeper benchmarks due to TraceMonkey merge

Status: RESOLVED INVALID

(Core :: JavaScript Engine, defect)

Description

[Alexandre Folle de Menezes]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.2) Gecko/20100115 Firefox/3.6 (.NET CLR 3.5.30729)
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.3a1pre) Gecko/20100203 Minefield/3.7a1pre (.NET CLR 3.5.30729)

I usually check the performance of the browser each time there's a TraceMonkey merge.

The nightly from 20100203 had a ~4% drop in performance (from 3506 to 3360) on the Peacekeeper benchmark, compared to 20100201 (I couldn't find 20100202).

Between the 2010-02-01 nightly build and the 2010-02-03 nightly builds, the result from the Peacekeeper regressed slightly, from 3506 to 3360 (~4%).  I couldn't find the build from 2010-02-02 nightly.

Detailed results:

Test                20100201  20100203
Rendering           2876      2826 (-1.7%)
Social Networking   3319      3165 (-4.9%)
Complex Graphics    5924      5587 (-6.0%)
Data                6711      6384 (-5.1%)
DOM Operations      2853      2722 (-4.8%)
Text Parsing        2902      2855 (-1.6%)

Reproducible: Always

Steps to Reproduce:
Just go to the URL and run the tests.

Actual Results:  
The results are ~5% lower than the build from feb/01.

Expected Results:  
Results in line with previous builds.

Comment 1

[Robert Sayre]

That's strange. Waldo was checking PeaceKeeper perf prior to this merge.

Comment 2

[Alexandre Folle de Menezes]

Well, this is not the first time there's a regression in this test, see bug #540985.

Comment 3

[Till Schneidereit [:till]]

Not relevant anymore with the removal of TM. Also, we're *a lot* faster on that and pretty much all other benchmarks then we were back then.