1435669 - 0.68% installer size (windows2012-32) regression on push 2b890d81a1e3ebc4872958022c8d425d5689b31d (Tue Jan 23 2018)

Status: RESOLVED WONTFIX

(Core :: DOM: Animation, defect)

Description

[Ionuț Goldan [:igoldan]]

We have detected a build metrics regression from push:

https://hg.mozilla.org/integration/autoland/pushloghtml?changeset=2b890d81a1e3ebc4872958022c8d425d5689b31d

As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

  1%  installer size windows2012-32 pgo      57,500,038.67 -> 57,888,660.83


You can find links to graphs and comparison views for each of the above tests at: https://treeherder.mozilla.org/perf.html#/alerts?id=11400

On the page above you can see an alert for each affected platform as well as a link to a graph showing the history of scores for this test. There is also a link to a treeherder page showing the jobs in a pushlog format.

To learn more about the regressing test(s), please see: https://developer.mozilla.org/en-US/docs/Mozilla/Performance/Automated_Performance_Testing_and_Sheriffing/Build_Metrics

Comment 1

[Ionuț Goldan [:igoldan]]

:hiro I filed this bug so we can properly track this Windows specific regression. Please use this bug for any fix you plan.

Comment 2

[Hiroyuki Ikezoe (:hiro)]

Here is a try with a change what I think it reduces the binary size somehow, it avoid the stupid thing on VC++ somehow.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=a56e23d9234e6105ad6775ba42a894a408f3df51&selectedJob=160375452

https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&originalRevision=ef1fefe4c6d1&newProject=try&newRevision=a56e23d9234e6105ad6775ba42a894a408f3df51&framework=2&filter=installer%20size%20pgo%20windows2012-32

  57,769,400.00 	> 	57,757,407.00 

It reduced approximately 12,000 bytes?  It's pretty small..

Comment 3

[Hiroyuki Ikezoe (:hiro)]

Oh, and I realized now this size increase happened only on PGO build.

https://treeherder.mozilla.org/perf.html#/compare?originalProject=autoland&originalRevision=499ac9e0f943&newProject=autoland&newRevision=2b890d81a1e3ebc4872958022c8d425d5689b31d&framework=2&filter=installer%20size%20windows2012

Comment 4

[Hiroyuki Ikezoe (:hiro)]

I did try another change that includes KeyframeUtils.cpp in KeyframeEffectReadOnly.cpp!

https://treeherder.mozilla.org/#/jobs?repo=try&revision=b563c36654dbbeef7ccb03b5513eefbff1673cb4&selectedJob=160382214

 installer size pgo: 57,763,413.00

I am almost inclined leave this.  PGO on Win32 is something broken.  Given that the template will be dropped once we stop building old style system (bug 1395112 and we have already build-time option for that (bug 1430014)), and this size increase should disappear at that time, I'd like to leave this problem as it is.

:igoldan, what do you think?

Comment 5

[Bobby Holley (:bholley)]

It does seem likely to me that the regression here is due to adding a separate monomorphization of code for a different stylecontext specialization. If that's the case, it's probably ok to leave until we drop the old style system, but it also seems like a really huge increase (almost the codesize weight of the entire gecko style system). So ideally we'd understand what's going on here. The answer may be "PGO works in mysterious ways", especially if we don't see it in other platforms.

Comment 6

[Ionuț Goldan [:igoldan]]

(In reply to Hiroyuki Ikezoe (:hiro) from comment #4)
> I did try another change that includes KeyframeUtils.cpp in
> KeyframeEffectReadOnly.cpp!
> 
> https://treeherder.mozilla.org/#/
> jobs?repo=try&revision=b563c36654dbbeef7ccb03b5513eefbff1673cb4&selectedJob=1
> 60382214
> 
>  installer size pgo: 57,763,413.00
> 
> I am almost inclined leave this.  PGO on Win32 is something broken.  Given
> that the template will be dropped once we stop building old style system
> (bug 1395112 and we have already build-time option for that (bug 1430014)),
> and this size increase should disappear at that time, I'd like to leave this
> problem as it is.
> 
> :igoldan, what do you think?

Sounds good to me.

Comment 7

[Hiroyuki Ikezoe (:hiro)]

For references, I did another try which has a change that means mostly backing out bug 1432104.

https://treeherder.mozilla.org/perf.html#/compare?originalProject=mozilla-central&originalRevision=0d806b3230fe&newProject=try&newRevision=183b347f31a0a629f23c4d8a0ee3ca346790cbe3&framework=2&filter=windows2012%20installer%20pgo

windows2012-32 pgo 57,768,976.00 	> 	57,766,834.00
windows2012-64 pgo 61,806,276.00 	< 	61,821,169.00 

It's mostly the same size.  The try was based on today's m-c 0d806b3230fe.  So, I guess the change for bug 1432104 increased binary size somehow, and later the binary size decreased somehow.  Or, backing out bug 1432104 is too late to restore the binary size?  I am not sure.