Open Bug 1562574 Opened 5 years ago Updated 2 years ago

When use mouse wheel to scroll through the docs.qq.com, the page will be significant slow.

Categories

(Core :: Graphics: Canvas2D, defect, P3)

All
Windows
defect

Tracking

()

Tracking Status
firefox-esr68 --- affected
firefox70 --- affected
firefox71 --- affected

People

(Reporter: xuyuehang, Unassigned)

References

()

Details

(Keywords: perf, Whiteboard: [sci-exclude])

Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:67.0) Gecko/20100101 Firefox/67.0 ID:20190619235627

After the page loaded, scroll page by mouse wheel, you will see page slower then before. The waterfall in devtool performance shows that the page frame rate suddenly drops below 10 when scrolling.

Sorry for the description not complete: I tested on two devices (both win10) including a blank profile can reproduce this problem. At the same time, it is good performance for my Firefox to scroll through other pages. This page is working fine under chrome.

https://perfht.ml/2J53Oh4

Most of the time is being spent in JS. The latency to the compositor is low so it's not an async scrolling issue. Given that it's a JS-implemented spreadsheet the scrolling is likely all happening in JS and speeding that would give us the biggest bang for the buck.

Component: Panning and Zooming → JavaScript Engine
Keywords: perf
OS: Unspecified → Windows 10
Hardware: Unspecified → x86_64

Most of the profile from comment 2 seems to be calling CanvasRenderingContext2D.

Component: JavaScript Engine → Canvas: 2D

The priority flag is not set for this bug.
:lsalzman, could you have a look please?

For more information, please visit auto_nag documentation.

Flags: needinfo?(lsalzman)

Not sure there's anything much that can be done here. We're basically just churning inside Direct2D doing stroke operations. Everything is already being accelerated.

Flags: needinfo?(lsalzman)
Priority: -- → P3
Whiteboard: [sci-exclude]

This is not a recent regression. I can reproduce it in Fx17.0 (with webpage break), Fx48, Fx60, Nightly 71, etc.

Significant performance difference for hardware acceleration on/off, lag only when the acceleration on.

Has Regression Range: --- → irrelevant
Has STR: --- → yes
OS: Windows 10 → Windows
Hardware: x86_64 → All
Version: unspecified → Trunk

Any thoughts for the new clue?

Flags: needinfo?(lsalzman)

The problem does not exist if gfx.direct2d.disabled = true.

It just looks like it is stuck trying to flush a lot of image draws on the GPU before stroking. I am not entirely sure why that would be slow.

Flags: needinfo?(lsalzman)

direct2d.disabled=false: https://perfht.ml/2PTAIFm
direct2d.disabled=true: https://perfht.ml/32pjSAS

Significant differences in Web Content - Category: DOM - d2d1.dll/d3d11.dll/igd[...].dll/etc.

Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.