1343558 - Compositor jank on xda-developers

Status: RESOLVED WORKSFORME

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

Description

[(no longer active)]

STR: Load https://www.xda-developers.com/deeply-integrated-progressive-web-apps-are-already-live-for-chrome-on-android/.

It seems to be 100% reproducible.  I tried several times and got the jank every time.

Here are some profiles:

https://perfht.ml/2me8y9z
https://perfht.ml/2me3Rg3
https://perfht.ml/2me70g5

Looks like the GLContextCGL::SwapBuffers() function in these profiles takes about 2.5 seconds or so easily.

Comment 1

[(no longer active)]

Same stack <https://perfht.ml/2mLpstz> on https://events.google.com/io/.

Comment 2

[Markus Stange [:mstange]]

Looking at these profiles again, I now have less confidence that bug 1265824 is going to help with this. Bug 1265824 is going to help with slow layer transactions (the red bars in the compositor thread), but probably not all that much with slow SwapBuffers (the blue bars).
These profiles look like we're just overwhelming the GPU in general.

Interestingly, on my new Macbook Pro I can't reproduce this bug. This machine has a rather powerful GPU though (Radeon Pro 460).

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

I don't see this on my Windows machine; Anthony, do you have access to a Windows machine (or few) to see what this looks like on them?

Comment 4

[(no longer active)]

This is another Mac specific issue.

Comment 5

[u279076]

I can't reproduce any jank with Nightly on my MacBook Pro (Retina, 13-inch, Late 2013) running MacOS Sierra 10.12.4 (16E195). A hardware profile would be useful in this case.

Comment 6

[Milan Sreckovic [:milan] (needinfo for best results)]

We're looking at bug 1342209 first.

Comment 7

[Milan Sreckovic [:milan] (needinfo for best results)]

Mac only, priority should be reconsidered.

Comment 8

[Naveed Ihsanullah [:naveed]]

Parent bug is qf:p1. This should be revisited then

Comment 9

[Gregory Pappas [:gregp]]

Unable to reproduce.