1221842 - Don't reallocate display item geometry if nothing changed

Status: RESOLVED FIXED

(Core :: Layout, defect)

Description

[Matt Woodrow (:mattwoodrow)]

Attachment: only-allocate-geometry-if-changed

Currently we always allocate a new nsDisplayItemGeometry when processing a display item, even if nothing actually changed.

If the invalidation region is empty, then nothing has changed, so we can just reuse the old geometry and make life easier for the memory allocator.

Comment 1

[Matt Woodrow (:mattwoodrow)]

We could take this a step further, and update existing geometry in-place when there are changes, but this is a rarer common case and a lot more work.

Comment 2

[Timothy Nikkel (:tnikkel)]

Does this work with nsImageGeometryMixin where we update the draw result sometimes but don't mark anything as invalid?

Comment 3

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/b456daa0503f

Comment 4

[Timothy Nikkel (:tnikkel)]

ni for comment 2

Comment 5

[Nigel Babu [:nigelb]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/e8b7b5321acddf4316dacfd521400fcef9c3ce7b
Backed out changeset b456daa0503f (bug 1221842) for B2G ICS Emulator Opt R7 bustage on a CLOSED TREE

Comment 6

[Matt Woodrow (:mattwoodrow)]

(In reply to Timothy Nikkel (:tn) from comment #2)
> Does this work with nsImageGeometryMixin where we update the draw result
> sometimes but don't mark anything as invalid?

Yeah, I believe this is fine.

We update the draw result on the stored geometry object, it shouldn't matter if this got freshly allocated during DLBI or was retained from the previous cycle.

Comment 7

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/230f12b4303c

Comment 8

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/230f12b4303c

Comment 9

[William Lachance (:wlach)]

This resulted in another nice improvement in the tp5o_scroll Talos benchmark:

https://treeherder.mozilla.org/perf.html#/alerts?id=2014&framework=1&hideImprovements=0