1371157 - cache results of system font lookups on Windows

Status: RESOLVED FIXED

(Core :: Widget: Win32, defect)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Profiling bug 1118086 on Windows shows that about 8% of the time in the big pause in the parent process to render a large select is time spent in mozilla::LookAndFeel::GetFont.  See the profile in https://perfht.ml/2qYJjqQ .

(Contrast with Linux, where time was spent in GetWidgetPadding / GetWidgetBorder; see bug 1367576.)

Caching the results of these repeated system font lookups should be a performance win, and should be pretty straightforward.

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=84b0b66b3f46ee147a75f44d9a5dbfb7485d15f9&group_state=expanded

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: Use override keyword in windows nsLookAndFeel header

MozReview-Commit-ID: 8bwmXmBFZB4

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Attachment: Cache results of Windows system font lookups

I haven't really tested that this fixes the performance problem observed
in a profile without the patch because the steps to use the Gecko
Profiler on local builds on Windows are rather complicated.

MozReview-Commit-ID: FmGFs2Cvquv

Comment 4

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

(The instructions for the "rather complicated" bit are: https://developer.mozilla.org/en-US/docs/Mozilla/Performance/Profiling_with_the_Built-in_Profiler_and_Local_Symbols_on_Windows , for what it's worth.)

Comment 5

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/aa3557e8515c8a8a1cbab8f5a5614476cc68ffe6
Bug 1371157 - Use override keyword in windows nsLookAndFeel header.  r=jimm

https://hg.mozilla.org/integration/mozilla-inbound/rev/70c026121670bd42e43bb87334fec1fd848b77d9
Bug 1371157 - Cache results of Windows system font lookups.  r=jimm

Comment 6

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/aa3557e8515c
https://hg.mozilla.org/mozilla-central/rev/70c026121670

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I verified that it fixed what I expected by reprofiling in today's nightly.

In today's nightly, 7ms are spent inside of ComputeFontData: https://perfht.ml/2svcp5g

Whereas in the profile from comment 0, that was 172ms: https://perfht.ml/2suzJAk of which 159ms was spent inside nsRuleNode::ComputeSystemFont (compared to 0 in today's nightly).