1983041 - Google Meet high sync IPC ReadPixels with the use of the background image feature

Status: RESOLVED WORKSFORME

(Core :: Graphics: Canvas2D, defect)

Description

[Jim Mathies [:jimm]]

https://share.firefox.dev/3UsiPO0

STR:

1. In Meet, enable the background image feature on your video. You may have to connect to meeting to set it up.
2. connect to a meeting. This can be a room with just ou as a participant, with your video in the panel or framed view on the screen.
3. wait a few minutes

results:
heavy fan use on this 2020 mac laptop

Some feedback from Andreas on Matrix -

The content process main thread is pretty hogged (only ~4% idle), most likely with video filter stuff (the message event is what their worker running a timer passes to the content process to drive the filter). Per the flame graph and inverted call tree it's mainly (35%) PWebGL::Msg_ReadPixels so sync gpu readbacks. Most of that time should be the CPU idling while waiting for it to complete, though, so fans spinning ought to be something else. The remaining 60% is pretty scattered, mainly wasm, then js and GC. The CanvasRenderer thread in the parent is where the readbacks happen. That thread is still ~60% idle, ~25% waiting for kernel messages (GPU?) and the remaining ~15% is work. Yeah, turning off the video filter should do a lot.

Comment 1

[Jim Mathies [:jimm]]

Attachment: about support

Comment 2

[Markus Stange [:mstange]]

I agree with what Andreas said - the cause is the fact that Google Meet calls ReadPixels to compute the video filter. I don't see anything particularly inefficient on the Firefox side in the profile - is performance better in Chrome on this machine?

Comment 3

[Jim Mathies [:jimm]]

Unfortunately Chrome displays no run up in cpu usage, and the device fans don't kick in after a few minutes in the video waiting room. So clearly Fx is struggling here compared to Chrome.

Comment 4

[BugBot [:suhaib / :marco/ :calixte]]

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

For more information, please visit BugBot documentation.

Comment 5

[Lee Salzman [:lsalzman]]

Do we have an equivalent profile from Chrome to verify among other things, if Zoom is even running the same code for Firefox and Chrome?