1838709 - 3,367,380 byte (3.6%) regression in APK size from -O2 compiler optimizations

Status: RESOLVED FIXED

(Firefox for Android :: Performance, defect, P1)

Description

[Chris Peterson [:cpeterson]]

Attachment: Perfherder_screenshot.png

Regression on Perfherder:

https://treeherder.mozilla.org/perfherder/graphs?highlightAlerts=1&highlightChangelogData=1&highlightCommonAlerts=0&series=firefox-android,4670243,1,2&series=firefox-android,4670244,1,2&series=firefox-android,4670245,1,2&series=firefox-android,4670246,1,2&timerange=2592000&zoom=1686614265147,1686870932513,83333055.87287581,102410547.31699347

Fenix regression range:

https://github.com/mozilla-mobile/firefox-android/compare/5fe9127947435c37948939b56b30f65a3b9c8d7c...348dc4badab80135f016076eb7c071d478afc96d

The GeckoView, Glean, and A-S updates are the suspects.

Comment 1

[Dianna Smith [:diannaS]]

The mozilla-central regression range based on the GV bumps is https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=b9e3497b939c3bdfee632c7297184b79cfeec9cc&tochange=6a3e558888107e82ba443f0ce749cd51f6e1b385

The A-S changes are here https://github.com/mozilla/application-services/compare/1e69fe5322b4c59027358167bbf978455d5534a6...6c4c65e8ce5808edd12260195e222ed988d780e5

Comment 2

[Jonathan Almeida [:jonalmeida]]

Marc, you had mentioned that this bump was probably because of the O2 optimization. Could you confirm two things:

1. That this is the case here.
2. We are still seeing a bump in APK size for x86 builds here, but I believe that the O2 optimization was only supposed to happen for arm64.

Comment 3

[Chris Peterson [:cpeterson]]

(In reply to Dianna Smith [:diannaS] from comment #1)

The mozilla-central regression range based on the GV bumps is https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=b9e3497b939c3bdfee632c7297184b79cfeec9cc&tochange=6a3e558888107e82ba443f0ce749cd51f6e1b385

Thanks, Dianna! The -O2 change (bug 1831935) is in that regression range and the timing aligns with this regression on June 15. We anticipated that -O2 change would increase the APK size.

(In reply to Jonathan Almeida [:jonalmeida] from comment #2)

2. We are still seeing a bump in APK size for x86 builds [here][0], but I believe that the O2 optimization was only supposed to happen for arm64.

Also, I thought we wanted to limit the -O2 change to Nightly builds (so we could see its impact without it affecting Beta or Release users yet), but the change doesn't check which channel is being built:

https://hg.mozilla.org/mozilla-central/rev/d2b14ac73d9a

Comment 4

[Marc Leclair [:mleclair]]

Yeah that is related to the O2 changes. Since its only in nightly right now I didn't land any other changes but there are ( I think) some changes to the compiler flag that we can add to reduce the impact of the O2 changes ( I think by like .8 MB or something like that).

Comment 5

[Chris Peterson [:cpeterson]]

[Tracking Requested - why for this release]:

(In reply to Marc Leclair [:mleclair] from comment #4)

Yeah that is related to the O2 changes. Since its only in nightly right now I didn't land any other changes but there are ( I think) some changes to the compiler flag that we can add to reduce the impact of the O2 changes ( I think by like .8 MB or something like that).

Marc, do you plan to revert your O2 changes before Nightly 116 merges to Beta (2023-07-03)? Or add a configure check to enable O2 if Nightly and Oz if Beta or Release? Either way, we don't want this 3 MB regression to ride to Beta or Release. :)

https://hg.mozilla.org/mozilla-central/rev/d2b14ac73d9a

Comment 6

[BugBot [:suhaib / :marco]]

The bug is marked as tracked for firefox116 (nightly). We have limited time to fix this, the soft freeze is in 6 days. However, the bug still isn't assigned and has low severity.

:amoya, could you please find an assignee and increase the severity for this tracked bug? Given that it is a regression and we know the cause, we could also simply backout the regressor. If you disagree with the tracking decision, please talk with the release managers.

For more information, please visit BugBot documentation.

Comment 8

[Marc Leclair [:mleclair]]

I'll get the numbers between nightly and release for pageload. As far as reverting the changes, I'll do that tomorrow so that it doesn't make it into release. However, I still think we should try to keep it in nightly afterwards to get a bit more data since this only landed roughly 10 days ago.

Comment 9

[Marc Leclair [:mleclair]]

This should now be fixed by bug 1840494 which will only build with O2 on nightly and Oz on all other variants.