976877 - 2 FPS scrolling http://davidad.github.io/blog/2014/02/25/overkilling-the-8-queens-problem/

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Benoit Girard (:BenWa)]

STR:
Scroll http://davidad.github.io/blog/2014/02/25/overkilling-the-8-queens-problem/

Seeing about 500ms. Most of the time is in 'argb32_image_mark_argb32'.

Comment 1

[Mayank Bansal]

same as scrolling this ?

https://brendaneich.com/2012/10/harmony-of-dreams-come-true/

Comment 2

[Markus Stange [:mstange]]

There's a CSS rule that sets opacity:0.99 on all divs. Deactivating that improves scrolling a lot.

Comment 3

[Matt Woodrow (:mattwoodrow)]

(In reply to Markus Stange [:mstange] from comment #2)
> There's a CSS rule that sets opacity:0.99 on all divs.

Some men just want to watch the world burn.

Comment 4

[Matt Woodrow (:mattwoodrow)]

Attachment: Don't simplify regions

This page exposed a few invalidation bugs. Bug 977418 is probably the solution for best performance though.


This is hurting us pretty badly on this page, and we have fast region code now so it shouldn't matter.

Comment 5

[Matt Woodrow (:mattwoodrow)]

Attachment: Pass layer state into build layer

ContainerState frequently decides to ignore the result of GetLayerState and decide something else. We should pass this new value down so that further down we don't have to call GetLayerState again (and potentially get the wrong value).

Comment 6

[Matt Woodrow (:mattwoodrow)]

Attachment: Don't build optimized layers in an inactive layer tree

If we scroll content off the page so that a layer is now only a solid color, then we're switching to a ColorLayer. LayerTreeInvalidation spots the new layer and invalidates the entire area, which is pretty sad for an inactive layer. Let's not do that.

Comment 7

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 8382768 [details] [diff] [review]
Pass layer state into build layer

Review of attachment 8382768 [details] [diff] [review]:
-----------------------------------------------------------------

::: layout/base/FrameLayerBuilder.cpp
@@ +3058,5 @@
>      offset = nsIntPoint(
>          NS_lround(NSAppUnitsToDoublePixels(appUnitOffset.x, appUnitsPerDevPixel)*aIncomingScale.mXScale),
>          NS_lround(NSAppUnitsToDoublePixels(appUnitOffset.y, appUnitsPerDevPixel)*aIncomingScale.mYScale));
>    }
> +  transform = transform * gfx3DMatrix::Translation(offset.x, offset.y, 0);

What's this change doing here?

Comment 8

[Matt Woodrow (:mattwoodrow)]

(In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #7)
> Comment on attachment 8382768 [details] [diff] [review]
> Pass layer state into build layer
> 
> Review of attachment 8382768 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> ::: layout/base/FrameLayerBuilder.cpp
> @@ +3058,5 @@
> >      offset = nsIntPoint(
> >          NS_lround(NSAppUnitsToDoublePixels(appUnitOffset.x, appUnitsPerDevPixel)*aIncomingScale.mXScale),
> >          NS_lround(NSAppUnitsToDoublePixels(appUnitOffset.y, appUnitsPerDevPixel)*aIncomingScale.mYScale));
> >    }
> > +  transform = transform * gfx3DMatrix::Translation(offset.x, offset.y, 0);
> 
> What's this change doing here?

Cancelling out the change to ContainerLayerParameters where we now include the parent offset. But I debugged this further, and this change doesn't actually fix anything properly.

It was just changing coordinates slightly so we were no longer losing to a different set of SimplifyOutward calls, for this page at least.

Comment 9

[Matt Woodrow (:mattwoodrow)]

Attachment: Don't simplify regions v2

Don't simplify in even more places.

These just keep failing and simplify outwards to the entire region, and we repaint way too much.

Comment 10

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/4e0303f0aebc
https://hg.mozilla.org/integration/mozilla-inbound/rev/3637434788e5

Comment 11

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/4e0303f0aebc
https://hg.mozilla.org/mozilla-central/rev/3637434788e5