1130545 - Increase tile size on OS X from 256x256 to something larger

Status: RESOLVED DUPLICATE of bug 1188995

(Core :: Graphics: Layers, defect)

Description

[George Wright (:gw280) (needinfo me!)]

We are currently using 256x256 tiles which can result in a large number of tiles being used for visible content (~150). This is suboptimal.

See also bug 1036682

Comment 1

[Boris Zbarsky [:bzbarsky]]

Fwiw, it's 150 per window.  If the user has a number of windows open, the number of tiles gets pretty big very quickly...

Comment 2

[Milan Sreckovic [:milan] (needinfo for best results)]

George, can you record all the conversations we had on this topic, and what the potential downside is to increasing the size?  We also covered whether we should somehow account for the screen resolution when choosing the size, or use non-square tiles.

Comment 3

[George Wright (:gw280) (needinfo me!)]

The main issue as far as I'm aware was that we use tiles on OS X for non-scrollable content, so if we increase the tile size we use up a lot of memory needlessly. We're already using a lot of memory for things like the scrollbar it seems, this just makes it worse.

I think ideally we want to move to a hybrid system where non-scrollable content uses non-tiled layers, and scrollable uses tiled. Markus had some reservations about that though.

We could also possibly look into using tiles for non-scrollable content, and setting the tile size to the size of that content.

Comment 4

[Markus Stange [:mstange]]

(In reply to George Wright (:gw280) from comment #3)
> I think ideally we want to move to a hybrid system where non-scrollable
> content uses non-tiled layers, and scrollable uses tiled. Markus had some
> reservations about that though.

Bug 1112193 explains the problem with that and how we should address it. In short: We need a new ContentClient implementation that is not tiled, not ContentClientIncremental, and doesn't require using sync transactions.

Comment 5

[Milan Sreckovic [:milan] (needinfo for best results)]

I know it's a soft dependency, but I'd say it's still useful to consider this one being blocked by the bug 1112193 Markus mentions above.

Comment 6

[Markus Stange [:mstange]]

This was done in bug 1188995.