1517940 - High CPU, bad performance on "Humble Book Bundle: Python 2019 by Packt" (clipping perf)

Status: RESOLVED WONTFIX

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

Description

[Asif Youssuff]

Came across this report on reddit - https://www.reddit.com/r/firefox/comments/abjrtr/humble_bundles_scrolling_background_imagine_makes/

Visit: https://www.humblebundle.com/books/python-packt-2019-books

Browsed to this on WR Linux and it takes up 100% of a core on my i7-7700hq. 

Performance profile: https://perfht.ml/2C2gqkl

Comment 1

[Jeff Muizelaar [:jrmuizel]]

I see 11ms GPU times on this page.

Comment 2

[Jeff Muizelaar [:jrmuizel]]

A bunch of time is being spent in C_Clip. Kats do you think this could be having the same problem you're seeing on the css 3d shooter page?

Comment 3

[Kartikaya Gupta (email:[e-mail])]

It looks like it could be the same problem, yeah. This page also has a perspective-transformed element (the animated background thingy at the top of the page) and the C_Clip times are high only when that is in the visible part of the viewport. That matches what I was seeing in bug 1504590 and it's likely the same root problem.

Comment 4

[Dzmitry Malyshau [:kvark]]

There isn't any obvious abnormality detected in a GPU capture. We fill out a full-screen surface with those tiles (non WR tiles). Then we draw a full-screen quad (transformed) with this texture. Then, on top of it, we draw a full-screen linear gradient.

Comment 5

[Sotaro Ikeda [:sotaro]]

(In reply to Asif Youssuff from comment #0)

Came across this report on reddit -
https://www.reddit.com/r/firefox/comments/abjrtr/
humble_bundles_scrolling_background_imagine_makes/

Visit: https://www.humblebundle.com/books/python-packt-2019-books

It seems that web page was changed and the problem does not happen.

Comment 6

[Asif Youssuff]

This page seems to have the same bad effect: https://www.humblebundle.com/games/double-fine-presents

Comment 7

[Jeff Muizelaar [:jrmuizel]]

I tried this and it didn't seem that bad to me. Is it still bad for you with picture-caching turned on?

Comment 8

[Asif Youssuff]

Tried to look into this and had to open Bug 1521205 - ran into a regression.

Comment 9

[gwarser]

I don't know if this will be useful, because I'm on old hardware q6600, gf8600gt, (driver340.107) on Manjaro KDE

Clean Nightly Build ID 20190120094340 with WR disabled saturates 2 out of 4 CPU cores, animation jams/jitters:

https://perfht.ml/2HmnZs5

After restart with WR on, cpu used - just above 100% of one core, but animation looks like 5 fps, plus my mouse cursor is very laggy on whole window and moves like 5 fps too.

https://perfht.ml/2RZsP2y - ~10s without touching anything and then mouse movements

For comparison Chromium 71 don't even uses full one core of cpu and animation is butter-smooth.

To eliminate such problems I use this user script which disables all transitions and animations.

Comment 10

[Asif Youssuff]

Testing again with the latest build.

CPU usage is still well above 100% across a couple of processes. Visually, everything looks fine, but I can feel the CPU heat up on my lap - with just Firefox and this page open.

Profile here: https://perfht.ml/2W4SINA but I fear it is not very interesting.

Comment 11

[Jeff Muizelaar [:jrmuizel]]

While bad, I don't think this needs to block our MVP.

Comment 12

[Nicolas Silva [:nical]]

Either the page or WebRender has changed, but this seems to run fine now.