Open
Bug 1421839
Opened 7 years ago
Updated 3 years ago
Firefox Quantum appears to struggle with a combination of the CSS filters and other element's movements.
Categories
(Core :: Graphics, enhancement, P3)
Core
Graphics
Tracking
()
NEW
| Tracking | Status | |
|---|---|---|
| firefox59 | --- | affected |
People
(Reporter: asa, Unassigned)
Details
(Whiteboard: [gfx-noted])
The test case is at https://codepen.io/michaelpumo/full/mqQWNj/
A video comparison on Mac is at https://streamable.com/gr3iw
The test setup was:
MacOS High Sierra
Macbook Pro 15", 2015
2.8 GHz Intel Core i7
16 GB RAM
Intel HD Graphics 630
Comment 1•7 years ago
|
||
Here's a profile of me hovering and then un-hovering the CodePen, on Linux Nightly:
https://perfht.ml/2BvpyvV
Looks like we've got some very-long rasterize operations (~450ms).
It's the filter-changes that are slow. (The profile has 98.5% of each long-rasterize-operation being spent in nsFilterInstance::PaintFilteredFrame -- and if I tweak the testcase to remove all of the transitions except for the filter one, it's still janky.)
Markus, should we be doing the filter as an operation on a layer here, or something like that? Or are there other obvious places where we're falling over here?
Component: CSS Parsing and Computation → Graphics
Flags: needinfo?(mstange)
Comment 2•7 years ago
|
||
Yes, moving this filter to the compositor would help a great deal, because then the filter operation would run on the GPU (at least on Mac). However, adding compositor support for filters is a bunch of work, and I'd rather invest that work into WebRender. WebRender already does pretty well on this testcase.
Bug 869496 is the bug about adding filter support to the compositor in the non-WebRender world.
Flags: needinfo?(mstange)
Updated•7 years ago
|
Whiteboard: [gfx-noted]
Updated•7 years ago
|
Priority: -- → P3
Updated•3 years ago
|
Severity: normal → S3
You need to log in
before you can comment on or make changes to this bug.
Description
•