Markus Stange [:mstange] Assignee
	Description • 10 years ago

When a container layer requires an intermediate surface, the surface is sized to its visible region. However, in some cases container layers lie about their visible region; the cases that immediately come to my mind are:
 - display ports and
 - prerendered transforms.

In these cases we ask the GPU to do a lot of work for stuff that won't be visible in the end.

	Markus Stange [:mstange] Assignee
	Comment 1 • 10 years ago

One way to fix this would be to have a "compute visibility" pass on the Compositor side prior to every composite that sets the shadow visible region to what will actually be visible of this layer in the current frame, taking async transforms into account. Then we can keep the logic that sizes the intermediate surface to the visible region bounds. We just need to make sure that it uses the shadow visible region if available.

	Markus Stange [:mstange] Assignee
	Comment 2 • 10 years ago

The combination of display ports and intermediate container layer surfaces can cause the fill ratio number to go up considerably.

	Markus Stange [:mstange] Assignee
	Comment 3 • 9 years ago

Bug 1021845 - Before compositing, clip the visible region of a layer to the layer's clip rect. r?mattwoodrow

	Markus Stange [:mstange] Assignee
	Comment 4 • 9 years ago

This patch is very similar to the second part in bug 1136766 (which you already reviewed), but it's not based on the first patch from that bug (because that caused performance regressions) and it's not limited to integer translations.

	Matt Woodrow (:mattwoodrow)
	Comment 5 • 9 years ago

Comment on attachment 8664842 [details]
MozReview Request: Bug 1021845 - Before compositing, clip the visible region of a layer to the layer's clip rect. r=mattwoodrow

https://reviewboard.mozilla.org/r/20039/#review18061

	Pulsebot
	Comment 6 • 9 years ago

https://hg.mozilla.org/integration/mozilla-inbound/rev/1da385d6c46b

	Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)
	Comment 7 • 9 years ago

Backed out for b2g r3 and r7 permafails: https://treeherder.mozilla.org/logviewer.html#?job_id=14641754&repo=mozilla-inbound

https://hg.mozilla.org/integration/mozilla-inbound/rev/e22824771c7f

	Joel Maher ( :jmaher ) (UTC -8)
	Comment 8 • 9 years ago

as a note, this caused a very large win for tcheck2 on android:
http://graphs.mozilla.org/graph.html#tests=[[201,64,29],[201,63,29]]&sel=none&displayrange=7&datatype=geo

	Markus Stange [:mstange] Assignee
	Comment 9 • 9 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=0758fb8b52a7

	Benoit Girard (:BenWa)
	Comment 10 • 9 years ago

Note that I added code to retain intermediate surfaces. I skimmed the patch but it wasn't clear if it interfered with that or not.

	Markus Stange [:mstange] Assignee
	Comment 11 • 9 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=2cc806364b57

	Markus Stange [:mstange] Assignee
	Comment 12 • 9 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=3613f9fdb6a7

	Markus Stange [:mstange] Assignee
	Comment 13 • 9 years ago

(In reply to Benoit Girard (:BenWa) from comment #10)
> Note that I added code to retain intermediate surfaces. I skimmed the patch
> but it wasn't clear if it interfered with that or not.

I don't think it will. Whenever the surface rect moves or changes size, we reallocate the surface.

	Kartikaya Gupta (email:kats@mozilla.staktrace.com)
	Comment 14 • 9 years ago

I assume you didn't mean to clear that dependency.

	Markus Stange [:mstange] Assignee
	Comment 15 • 9 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=ddcf88690dc3

	Markus Stange [:mstange] Assignee
	Comment 16 • 9 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=45740e816e15

	Markus Stange [:mstange] Assignee
	Comment 17 • 9 years ago

Comment on attachment 8664842 [details]
MozReview Request: Bug 1021845 - Before compositing, clip the visible region of a layer to the layer's clip rect. r=mattwoodrow

Review request updated; see interdiff: https://reviewboard.mozilla.org/r/20039/diff/1-2/

	Markus Stange [:mstange] Assignee
	Comment 18 • 9 years ago

Bug 1021845 - Don't skip checkerboarding layers during compositing, even if the layer's visible region is empty. r?botond

	Botond Ballo [:botond]
	Comment 19 • 9 years ago

https://reviewboard.mozilla.org/r/20039/#review24043

::: gfx/layers/composite/LayerManagerComposite.cpp:209
(Diff revision 2)
> +IntersectMaybeRects(const Maybe<RectType>& aRect1, const Maybe<RectType>& aRect2)

In bug 1168263, I added an IntersectMaybeRects() function to AsyncCompositionManager.cpp. We may want to consider moving it to a shared header and reusing it.

	Botond Ballo [:botond]
	Comment 20 • 9 years ago

Comment on attachment 8693528 [details]
MozReview Request: Bug 1021845 - Don't skip checkerboarding layers during compositing, even if the layer's visible region is empty. r?botond

https://reviewboard.mozilla.org/r/26207/#review24045

	Markus Stange [:mstange] Assignee
	Comment 21 • 9 years ago

https://treeherder.mozilla.org/#/jobs?repo=try&revision=d3381c4d4908

	Markus Stange [:mstange] Assignee
	Comment 22 • 9 years ago

https://hg.mozilla.org/integration/mozilla-inbound/rev/10fb062fb1ae906a0ad43555ee0f62dfe7ec667f
Bug 1021845 - Before compositing, clip the visible region of a layer to the layer's clip rect. r=mattwoodrow

https://hg.mozilla.org/integration/mozilla-inbound/rev/fc0ac31d14eb4b218df818e164bf9de897cf561a
Bug 1021845 - Don't skip checkerboarding layers during compositing, even if the layer's visible region is empty. r=botond

	Carsten Book [:Tomcat]
	Comment 23 • 9 years ago
bugherder

https://hg.mozilla.org/mozilla-central/rev/10fb062fb1ae
https://hg.mozilla.org/mozilla-central/rev/fc0ac31d14eb

	Ryan VanderMeulen [:RyanVM]
	Comment 25 • 9 years ago

This was backed out from Firefox 45.

https://hg.mozilla.org/releases/mozilla-aurora/rev/64ec448f156d