1526970 - nsIFrame::(Add|Remove)DisplayItem() is slow

Status: RESOLVED FIXED

(Core :: Web Painting, enhancement, P2)

Description

[Miko Mynttinen]

This shows up on profiles a lot. We could merge the frame property |nsIFrame::DisplayItems| with a member variable |nsIFrame::mDisplayItemData| that holds DisplayItemData, or swap them to speed up WebRender that does not use DisplayItemData.

Comment 1

[Miko Mynttinen]

Changing the frame property to member variable gives a massive performance boost in various tests: https://treeherder.mozilla.org/perfherder/compare?originalProject=try&originalRevision=a269eee28a93d4baaf366816c799df2262b4c755&newProject=try&newRevision=c86901b25aa5e814daf2f6a76d8f582e6a53c1c0&framework=1

Comment 2

[Miko Mynttinen]

Attachment: Bug 1526970 - Part 1: Store the list of display items the frame has as a member variable instead of frame property r=mattwoodrow

Comment 3

[Miko Mynttinen]

Attachment: Bug 1526970 - Part 2: Convert nsIFrame::mDisplayItemData to frame property r=mattwoodrow

Depends on D104059

Comment 4

[Miko Mynttinen]

Attachment: Bug 1526970 - Part 3: Remove nsPaintedDisplayItem::mDisplayItemData r=mattwoodrow

Depends on D106164

Comment 5

[Miko Mynttinen]

Attachment: Bug 1526970 - Part 4: Check the frame for display items rather than DisplayItemData or WebRenderFallbackData in InvalidateFrame() r=mattwoodrow

Depends on D106165

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/fed61f8125da
Part 1: Store the list of display items the frame has as a member variable instead of frame property r=mattwoodrow,emilio
https://hg.mozilla.org/integration/autoland/rev/d5fc307deac1
Part 2: Convert nsIFrame::mDisplayItemData to frame property r=mattwoodrow
https://hg.mozilla.org/integration/autoland/rev/822a22e21e7a
Part 3: Remove nsPaintedDisplayItem::mDisplayItemData r=mattwoodrow
https://hg.mozilla.org/integration/autoland/rev/d5b874ccd877
Part 4: Check the frame for display items rather than DisplayItemData or WebRenderFallbackData in InvalidateFrame() r=mattwoodrow

Comment 7

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/fed61f8125da
https://hg.mozilla.org/mozilla-central/rev/d5fc307deac1
https://hg.mozilla.org/mozilla-central/rev/822a22e21e7a
https://hg.mozilla.org/mozilla-central/rev/d5b874ccd877

Comment 8

[Alexandru Ionescu (needinfo me) [:alexandrui]]

== Change summary for alert #28981 (as of Tue, 02 Mar 2021 04:42:14 GMT) ==

Improvements:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
12%	tp5o_scroll		windows10-64-shippable-qr	e10s stylo webrender-sw	2.16 -> 1.91
6%	tp5o_scroll		linux1804-64-shippable	e10s stylo	1.71 -> 1.61
6%	tp5o_scroll		linux64-shippable	e10s stylo	1.41 -> 1.33
4%	tp5o_scroll		linux1804-64-shippable-qr	e10s stylo webrender	2.70 -> 2.58
4%	tp5o_scroll		linux1804-64-shippable-qr	e10s stylo webrender	2.67 -> 2.56
3%	tp5o_scroll		linux1804-64-shippable-qr	e10s stylo webrender	2.67 -> 2.58

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=28981

Comment 9

[Cosmin Sabou [:CosminS]]

Backed out as requested by miko for causing a perma-fail, backout link: https://hg.mozilla.org/integration/autoland/rev/2ed2319aa1d7eddd4448d88dc2bea5f3f013b2bc

Comment 10

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Backout merged: https://hg.mozilla.org/mozilla-central/rev/2ed2319aa1d7

Comment 11

[Miko Mynttinen]

The backed out part was additional optimization that was based on a wrong assumption. We can keep this closed.