Horrible Octane 2.0 performance for native Firefox x86-64 for Linux
Categories
(Core :: JavaScript Engine: JIT, defect, P5)
Tracking
()
Performance Impact | none |
People
(Reporter: aros, Unassigned)
References
(Blocks 2 open bugs)
Details
Attachments
(2 files)
Steps to reproduce:
Consider https://chromium.github.io/octane/
Firefox 100 x86-64, native official Mozilla Linux build: roughly 40K points
SeaMonkey 2.53.12 x86 running under Wine x86: roughly 60K points.
Actual results:
The native x86-64 version is running ~33% (!!) slower than the "emulated" x86 version of SeaMonkey 2.53.12.
Expected results:
Versions used:
SeaMonkey 2.53.12: https://archive.mozilla.org/pub/seamonkey/releases/2.53.12/win32/en-US/seamonkey-2.53.12.en-US.win32.installer.exe
Firefox 100: https://ftp.mozilla.org/pub/firefox/releases/100.0/linux-x86_64/en-US/firefox-100.0.tar.bz2
The test is 100% reproducible. In both situations I used new profiles with no changes or any extensions installed.
Reporter | ||
Comment 1•2 years ago
|
||
Updated•2 years ago
|
Comment 2•2 years ago
•
|
||
I managed to reproduce this issue on Ubuntu 20.04 on Firefox 100, 101 and 102.
Setting this as New, updating the flags and waiting for the developer's opinion about it.
Edit: fixed a typo in the spelling.
Comment 3•2 years ago
|
||
I suspect this is due to divergence in JIT features. No direct performance impact as this is not on an actual website.
Comment 4•2 years ago
|
||
“SeaMonkey 2.53.12 uses the same backend as Firefox and contains the relevant Firefox 60.8 security fixes.” — https://www.seamonkey-project.org/releases/seamonkey2.53.12/
This is likely to be the difference between IonBuilder + TypeInference versus WarpBuilder.
While our focus moved away from benchmarketting, this is a problem which should be addressed by WatchTower.
Updated•5 months ago
|
Updated•5 months ago
|
Description
•