Open Bug 1707325 Opened 4 years ago Updated 1 year ago

FF has very low FPS

Categories

(Core :: Graphics: Canvas2D, defect)

Firefox 89
x86_64
Windows 10
defect

Tracking

()

People

(Reporter: ezh, Unassigned, NeedInfo)

References

(Depends on 1 open bug)

Details

Attachments

(1 file)

  1. Open https://astrotourismpsb.rbc.ru/
  2. Place the cursor at the point in the right side and hold
  3. See what happens (and how slow)
  4. On the next screen holding left mouse key and moving should navigate around the Space
    In FF it is nearly impossible, FF is very unresponsive on the page, unusable at all.
    In Chrome and Co it is fine.

Compositing WebRender
Ryzen 7 4700U

Tomorrow will try on other PC's and notebooks I have.

The Bugbug bot thinks this bug should belong to the 'Core::Graphics: WebRender' component, and is moving the bug to that component. Please revert this change in case you think the bot is wrong.

Component: General → Graphics: WebRender
Product: Firefox → Core
Component: Graphics: WebRender → Canvas: 2D

Lee, anything we can/should do here?
Fwiw, I tested on NV GTX 1050 / Linux, and it runs fine.
The profile Mayank shared is also not terrible: composite times are 10-30ms. It's not the fastest, but it should be usable.

Severity: -- → S3
Flags: needinfo?(lsalzman)

Can you please attach your "about:support" information so we can get a better idea what's going on?

Flags: needinfo?(lsalzman) → needinfo?(ezh)
Attached file about.support.txt

CPU: Ryzen 5600X, 32Gb RAM

It is better, than on a notebook I mentioned before, but Edge is far more quicker and responsive.

Flags: needinfo?(ezh)

Video on the Ryzen 5600X PC: https://photos.app.goo.gl/2NjUo8TQXsZZmHhN7

JeffM, looks like he is using HW-WR with D2D canvas. Seems like from the profile that we are spending a lot of time in RenderDXGITextureHost locking the texture. I am not quite sure if that is the bottleneck, in so far as the canvas is doing a lot of rendering, moreso than WR itself, so I would expect WR to be waiting on it, but not quite sure overall why it might make everything so slow. Any ideas?

Flags: needinfo?(matt.woodrow)
Flags: needinfo?(jmuizelaar)
Flags: needinfo?(jgilbert)

Bob, could this have some relation to canvas remoting? I am not seeing a lot of overhead in the content process itself, so this is strange.

Flags: needinfo?(bobowencode)
Flags: needinfo?(matt.woodrow)
Flags: needinfo?(jgilbert)

Oops, only meant to NI JeffM.

Seeing a potentially similar issue in bug 1707336. Where we are spending a lot of time waiting on D2D canvas, in a canvas worker thread. This could potentially be different manifestations of the same problem.

See Also: → 1707336

(In reply to Lee Salzman [:lsalzman] from comment #8)

Bob, could this have some relation to canvas remoting? I am not seeing a lot of overhead in the content process itself, so this is strange.

I don't think so, with remote canvas turned off it waits for Direct2D in a similar way, just in the content process main thread.

Flags: needinfo?(bobowencode)

This is slow with D2d-canvas: https://share.firefox.dev/3Yr5VmG
But is fast with gpu-canvas: https://share.firefox.dev/3ynu42X

Depends on: gpu-canvas
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: