Bugzilla – Bug 1160994
Firefox very slow after update to version 72
Last modified: 2022-03-26 19:44:04 UTC
Created attachment 827577 [details]
After update to version 72 on tumbleweed, firefox has become painfully slow. The problem did not occur with version 71.
example 1: google maps is almost unusable.
example 2: the graphics tests 4,5,6,7 on https://web.basemark.com/
Most importantly: If I download binaries for Firefox 72 directly from https://www.mozilla.org, the problem does *not* occur, Firefox is as speedy as usual.
The problem occurs even when starting with a new, clean profile (deleting .mozilla in home directory).
The problem occurs on two different laptops.
about:support is attached (both for the version from the tumbleweed repo and for the version directly from mozilla).
I do not see this affect neither with TW nor with Leap.
Can you please also share your desktop environment?
GFX hardware seems to be Whiskey Lake 3x8 GT2?
And because I'm wondering about basemark. I have no comparison because I haven't used it before. What is your total score there?
I am on KDE (tumbleweed snapshot 20200113). Correct, Whiskey Lake 3x8 GT2 (Intel UHD graphics 620).
The second laptop on which the same problem occurs is also running KDE (tumbleweed) and also has Intel graphics (Iris Plus 640).
Basemark score for Firefox 72 (clean profile) from Tumbleweed repo: 270
Basemark score for Firefox 72 (clean profile) from Mozilla: 390
However, these scores might be misleading since not all tests seems similarly affected - it seems that only the graphics-intensive tests are affected. Example:
Test 7 (draw-call-test) runs 30 seconds on Mozilla's build and about 8 minutes on Tumbleweed's build. I could share a screen capture if needed.
As a test, I just installed firefox-esr 68.4.1 from the tumbleweed repo - esr is *not* affected!
PS: I just ran a diff on about:support, comparing Mozilla's and TW's build. Config seems exactly the same:
diff mozillabuild.txt twbuild.txt
< Update Channel: release
< dom.push.userAgentID: 0cb13890125c405ebfc0b752dd3fdf9c
> dom.push.userAgentID: 1a53a2af73374dfea303607f17ab448a
< security.sandbox.content.tempDirSuffix: b302fa80-775c-4ee1-8343-2a28d5c6f794
< security.sandbox.plugin.tempDirSuffix: e0120e53-f70b-484a-8d73-d1c0fd06a217
> security.sandbox.content.tempDirSuffix: 269201ae-9c87-46c2-a716-afb815eb7e1b
> security.sandbox.plugin.tempDirSuffix: 3e842a6a-a45d-494e-b431-51059e5cbd35
< App Locales: ["en-US"]
< Regional Preferences: ["en-US"]
< Default Locale: "en-US"
> App Locales: ["en-US","und"]
> Regional Preferences: ["en-US","und"]
> Default Locale: "und"
I can confirm a performance degradation related to (likely) graphics.
Compared 72.0.1 and 68.4.1 on my Leap 15.1 system:
- graphics from basemark moving faster
Just a very quick thought does not reveal larger differences in build and patches between both.
Obviously some typical fluctuation in patches being obsolete or rebased.
So the difference between openSUSE versions of ESR and regular are VERY small. Compared with upstream we use gcc instead of clang but as ESR is not (as much?) affected as regular it somehow think that it's a combination of Firefox core changes and probably toolchain.
As a first step I can switch from gcc to clang to see if that is related.
(In reply to Christoph Karrasch from comment #3)
> PS: I just ran a diff on about:support, comparing Mozilla's and TW's build.
> Config seems exactly the same:
since you have upstream available right now. Can you please compare about:buildconfig?
Created attachment 827630 [details]
Created attachment 827631 [details]
buildconfig, diff tw - upstream
Got a clang build of FF72.0.1 on my machine now.
I could not guarantee lab conditions in my benchmarks but what I had with web.basemark.com:
gcc build: 270
clang build: 301
Some anymations were visibly faster but 68.4.1 still had a higher score (built with gcc).
Also Christoph's findings:
> Basemark score for Firefox 72 (clean profile) from Tumbleweed repo: 270
> Basemark score for Firefox 72 (clean profile) from Mozilla: 390
are not close. I had 270 but I only reached 301 compared to 390 with clang.
I try to build and publish the clang build from here:
but OBS seems to have worker issues so huge packages like Firefox are waiting currently for ages to be built.
Once some build is available. I would need feedback from comparison on other people's machines.
> I try to build and publish the clang build from here:
> but OBS seems to have worker issues so huge packages like Firefox are
> waiting currently for ages to be built.
> Once some build is available. I would need feedback from comparison on other
> people's machines.
I just checked this repo, it seems that the x86_64 build for tumbleweed failed?
We are a couple of major versions ahead. Assuming his is fixed now. Reopen if needed with current data.