1620076 - [WebRender] Linux partial present

Status: RESOLVED FIXED

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

Description

[Jeff Muizelaar [:jrmuizel]]

This will make things better

Comment 1

[Robert Mader [:rmader]]

See also bug 1484812 for some pre-existing work on that. To make use of EGL_KHR_swap_buffers_with_damage on X11, we'll need to fix bug 788319 first however, which again depends on bug 1474281.

Comment 2

[Val Packett]

Attachment: Bug 1620076 - Partial compositing (damage) with EGL_EXT_buffer_age in WebRender

EGL requires the application to keep the front buffer fully consistent.
This means we have to draw the previous frame's damage as well.
(But we don't need to include it in the hint we're sending to the
system compositor via SwapBuffersWithDamage.)

Comment 3

[Val Packett]

Reassigned my webrender patch from bug 1484812 to here, since this is the bug for WR (please mention [WebRender] in the title!)

The patch works (on Wayland) but still has some condition that leads to leftover artifacts (some example cases: VAAPI video playback, some in-page popover things on some pages that happen on hover but not click..)

Comment 4

[Sotaro Ikeda [:sotaro]]

Created Bug 1625070 for partial present on X11.

Comment 5

[Val Packett]

Looks like D71139 (Bug 1611372) or something else that happened in the last ~month might've helped! From a quick attempt, I can't reproduce the artifacts.

Also, made the patch respect gfx::gfxVars::WebRenderMaxPartialPresentRects for a way to toggle it without rebuilding.

Comment 6

[Robert Mader [:rmader]]

(In reply to greg v [:myfreeweb] from comment #5)

Looks like D71139 (Bug 1611372) or something else that happened in the last ~month might've helped! From a quick attempt, I can't reproduce the artifacts.

Also, made the patch respect gfx::gfxVars::WebRenderMaxPartialPresentRects for a way to toggle it without rebuilding.

Nice! Just gave it a try and after setting gfx.webrender.max-partial-present-rects to a value bigger than 0 I could verify on GS/Mutter that partial damage works (though WR seems to invalidate much more than the basic layer atm)! Couldn't spot any issues either, but also tried very long. More reasons for me to get going on bug 788319 so it works for X11, too :)

Jeff, do you think you could have a look into D61062? At least if it goes in the right direction?

Comment 7

[Jeff Muizelaar [:jrmuizel]]

I've redirected that request to Glenn and Sotaro.

Comment 8

[Val Packett]

WR seems to invalidate much more than the basic layer atm

WR currently is hardcoded to merge all damage rects into one due to other platforms.

Also I might be forcing more frames to render than necessary, sometimes buffer age is 3 (wat??) and maybe it's fine to use the same method on these frames too.

Comment 10

[Sotaro Ikeda [:sotaro]]

:jgilbert, can you review https://phabricator.services.mozilla.com/D61062?

Comment 11

[Robert Mader [:rmader]]

As this got accepted from Glen (in a comment) and Jeff - can we push it? :)

Comment 12

[Robert Mader [:rmader]]

Minor bummer for the plan to use EGL_KHR_swap_buffers_with_damage on X11, too: it's not yet supported on X11 by mesa. Just opened a issue (1). I suppose driver support for partial damage will be a requirement to ship WR by default, at least when on battery. Maybe it's even a good marker - it will guarantee us a recent mesa version :)

Note: bug 1640048 implements a working X11 EGL implementation that, in combination with the patch from this bug, should bring us as far as we can get on X11 concerning compositor integration, i.e. partial damage and opaque regions. Everything more fancy will likely be Wayland only, see bug 1617498.

1: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3030

Comment 13

[Martin Stránský [:stransky] (ni? me)]

(In reply to Robert Mader [:rmader] from comment #11)

As this got accepted from Glen (in a comment) and Jeff - can we push it? :)

Yes, please set "Checkin needed" Tag at
https://phabricator.services.mozilla.com/differential/revision/edit/61062/

Comment 14

[Robert Mader [:rmader]]

(In reply to Martin Stránský [:stransky] from comment #13)

Yes, please set "Checkin needed" Tag at
https://phabricator.services.mozilla.com/differential/revision/edit/61062/

Unfortunately I don't have the permissions for that - so waiting for Greg or somebody else (you?) to do it :)

Comment 15

[Jan Alexander Steffens [:heftig]]

I did it. I assume I can do that because I'm part of the "bmo-editbugs-team".

Maybe you should apply for editbugs permissions? https://bugzilla.mozilla.org/page.cgi?id=get_permissions.html

Comment 16

[Robert Mader [:rmader]]

(In reply to Jan Alexander Steffens [:heftig] from comment #15)

I did it. I assume I can do that because I'm part of the "bmo-editbugs-team".

Maybe you should apply for editbugs permissions? https://bugzilla.mozilla.org/page.cgi?id=get_permissions.html

Thanks! And good point, just requested it :)

Comment 17

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/90998e8e1bdd
Partial compositing (damage) with EGL_EXT_buffer_age in WebRender r=jgilbert

Comment 18

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/90998e8e1bdd

Comment 19

[Robert Mader [:rmader]]

Thank at Greg and anybody involved! I just gave it a try and there's still one bit missing: in order to enable this, gfx.webrender.max-partial-present-rects has to get set to a non-default value (-1 or 64 or so). Is there anything blocking us from doing that? Sotaro, do you maybe know?

Comment 20

[Val Packett]

(In reply to Robert Mader [:rmader] from comment #19)

I just gave it a try and there's still one bit missing: in order to enable this, gfx.webrender.max-partial-present-rects has to get set to a non-default value (-1 or 64 or so). Is there anything blocking us from doing that?

I wasn't 100% confident that it's artifact-free so I didn't touch the default :)

In particular, gfx.webrender.debug.profiler is not very usable for me with partial present on.

In normal operation, it's fine: I can't reliably reproduce artifacts with popovers in pages or the URL thing on hover anymore, it's like 99.9% good, but I've seen a little artifact maybe a couple times in a few days.

Comment 21

[Darkspirit]

So far I haven't seen other issues than these two debug overlay bugs. If they don't indicate a severe problem, then it should be fine to enable partial present by default.

Comment 22

[Robert Mader [:rmader]]

I'd like to track as many bugs concerning this as possible in bug 1640710 - i.e. all bugs that are not Wayland/X11 specific but can e.g. reproduced on Windows with DirectComposition disabled (Win 7/8).

Comment 23

[Sotaro Ikeda [:sotaro]]

(In reply to Robert Mader [:rmader] from comment #19)

Is there anything blocking us from doing that? Sotaro, do you maybe know?

Except the 2 bugs, I do not know another bugs. I seems OK to enable partial present.

Comment 24

[Mel]

(In reply to Sotaro Ikeda [:sotaro] from comment #23)

(In reply to Robert Mader [:rmader] from comment #19)

Is there anything blocking us from doing that? Sotaro, do you maybe know?

Except the 2 bugs, I do not know another bugs. I seems OK to enable partial present.

https://bugzilla.mozilla.org/show_bug.cgi?id=1640858

Comment 25

[Robert Mader [:rmader]]

Site node: we just fixed a quite bad bug in Mutter that made partial damage much less effective. So if you're running Mutter, the effect of enabling gfx.webrender.max-partial-present-rects should be much bigger (when looking at the whole system) after the upgrade to 3.36.4 or 3.38. This of course also affects the default software rendering, which has a quite optimized partial damage implementation.

See https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1326