1663152 - MOZ_X11_EGL/proprietary Nvidia: WebGL 2 is not available: "WebGL 2 requires support for the following features: get_integer64_indexed"

Status: VERIFIED FIXED

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

Description

[Darkspirit]

Attachment: aboutsupport.txt

Gnome X11, Debian Testing, Nvidia GTX 1060, driver 450.57

Since bug 1650583 it's possible to use MOZ_X11_EGL=1 with proprietary Nvidia on Linux.

MOZ_X11_EGL=1 mozregression --launch 20200903151816 --pref gfx.webrender.all:true widget.dmabuf-webgl.enabled:false -a about:support -a http://webglsamples.org/WebGL2Samples/#draw_image_space

about:support:

WebGL 2 Driver Renderer: WebGL creation failed:

• WebGL 2 requires support for the following features:
  get_integer64_indexed

It works with GLX.

Comment 1

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

We can't use EGL with proprietary NVIDIA drivers.

Comment 2

[Hi-Angel]

(In reply to Martin Stránský [:stransky] from comment #1)

We can't use EGL with proprietary NVIDIA drivers.

Thanks, I'm just curious, could you elaborate on this or provide a link to a discussion. I searched over the #788319, but haven't found anything about that.

Comment 3

[Darkspirit]

(In reply to Martin Stránský [:stransky] from comment #1)

We can't use EGL with proprietary NVIDIA drivers.

I try to summarize and have some questions. Please complement if incomplete.
EGL didn't work with proprietary Nvidia because it doesn't allow to use GLX (to find X visual) and EGL at the same time (bug 1650583).
EGL/proprietary Nvidia worked by preventing parallel GLX usage, but it caused bug 1663003 (which showed bug 1479135 even to Mesa/Nvidia users with compositing window managers):

• The change had to be reverted with bug 1663003 comment 17 because of a Mesa bug that prevents transparency on EGL windows: "Use GLX to find X Visual for EGL as a workaround for https://gitlab.freedesktop.org/mesa/mesa/-/issues/149"
  • Work has resumed here: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/2376#note_623834
• Proprietary Nvidia is apparently back to the original bug?
  • Nvidia commenting on EGL transparency on X11: https://bugs.freedesktop.org/show_bug.cgi?id=67676#c17
    https://lists.x.org/archives/mesa-dev/2015-April/082140.html
  • EGL/X11/Proprietary Nvidia seems to be desired because it supports partial present (bug 1625070 comment 30) while GLX does not.
  • Questions:
    • Could either Mesa or Proprietary Nvidia be detected and GLX Visual only be used on Mesa (until the Mesa bug is fixed)? Then, only users of the proprietary Nvidia driver would encounter bug 1663003 (bug 1479135) when using MOZ_X11_EGL=1.
    • Could fixing bug 1479135 comment 44 solve the main problem for Nvidia? (In doubt, could the fix only be used on proprietary Nvidia?)
    • Could this missing WebGL feature be reported to Nvidia so they could fix it (if it's a problem on their side)? Or are there more severe problems?

Comment 4

[Darkspirit]

(In reply to Martin Stránský [:stransky] from comment #1)

We can't use EGL with proprietary NVIDIA drivers.

(Martin Stránský [:stransky] from bug 1650583comment #22)

When GLX Vsync source is created along EGL contexts, NVIDIA drivers refuse to make any EGL content current.
So disable GLX Vsync source creation when EGL context is used.

Just plugged my GTX1060 back onto the mainboard: It's much less bad than feared.
As GLX is apparently only used for Visual (bug 1663003 comment 17), but not for vsync, nothing is broken.
MOZ_X11_EGL=1 still works, there are no black borders. There is just this one WebGL2 problem.
Even EGL_KHR_swap_buffers_with_damage works! Nice! (Debian Testing, Gnome X11 on proprietary Nvidia driver 450.66)

Comment 5

[Darkspirit]

Attachment: aboutsupport_webgl2_works_when_requiring_OGL3.2_on_EGL.txt

I don't know why it's only a problem on EGL/proprietary Nvidia, but I've found the cause:

https://searchfox.org/mozilla-central/rev/1a973762afcbc5066f73f1508b0c846872fe3952/gfx/gl/GLContextFeatures.cpp#185

{"get_integer64_indexed",
GLVersion::GL3_2,
GLESVersion::ES3,
GLContext::Extension_None,
{GLContext::Extensions_End}},

This bug can be fixed by requiring OpenGL 3.2 instead of 3.1 on EGL in this line.
GLX already requires OpenGL 3.2.

Attached about:support (MOZ_X11_EGL=1 ./mach run) shows working WebGL2 on EGL/X11/Proprietary Nvidia.
Would it be okay to make this change? Shall I submit this one line patch via Phabricator or would you expect problems for other targets (should it be gated by MOZ_WIDGET_GTK)?

Background info: I don't know if it's related, but since bug 1474281 OGL is preferred over GLES when using EGL.

(Benefit: Proprietary Nvidia supports EGL_KHR_swap_buffers_with_damage (bug 1648799) when using MOZ_X11_EGL=1.)

Comment 6

[Kelsey Gilbert [:jgilbert]]

Weird, we expect the driver to give us 3.2 when we request 3.1, if 3.2 is supported! I suppose this might be due to the awkward two-stage compat/core profile introduction that added compat in 3.1, but only added core profiles in 3.2.
I think we should just ask for 3.2 instead of 3.1 here.
Good investigation!

Comment 7

[Kelsey Gilbert [:jgilbert]]

Attachment: Bug 1663152 - Ask for 3.2core+ if !REQUIRE_COMPAT_PROFILE on EGL.

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/016992fe00d7
Ask for 3.2core+ if !REQUIRE_COMPAT_PROFILE on EGL. r=lsalzman

Comment 9

[Dorel Luca [:dluca]]

Backed out changeset 016992fe00d7 (bug 1663152) for Browser-chrome failures in browser/base/content/test/performance/browser_startup_mainthreadio.js. CLOSED TREE

Log:
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=318146609&repo=autoland&lineNumber=3302

Push with failures:
https://treeherder.mozilla.org/#/jobs?repo=autoland&group_state=expanded&revision=016992fe00d74a98ecaa22b6071cf3bade84e83d

Backout:
https://hg.mozilla.org/integration/autoland/rev/48b695c0f3b9b8843de15f9b8ffe694c00f57724

Comment 10

[Darkspirit]

TEST-UNEXPECTED-FAIL | browser/base/content/test/performance/browser_startup_mainthreadio.js | unexpected stat on C:\Users\task_1602223743\AppData\Local\Mozilla\firefox before first paint -

This so far unexpected action only occured on "Windows 10 x64 WebRender opt".

2 months ago, there were similar previously unexpected actions and they have been allowed:

(Joel Maher ( :jmaher ) (UTC -0400) from bug 1658443 comment 0)

I am working to get our windows10 tests running on hardware (and fallback to the cloud when busy) in order to use our spare cpu time in the datacenter. most tests pass juts fine, but a few fail.

I can add exceptions to the files, is there more work that needs to be done before doing that?

(Joel Maher ( :jmaher ) (UTC -0400) from bug 1660582 comment 0)

in bug 1658443 I reported a set of file/IO access while running tests on windows10 hardware instead of VMs at AWS.

TEST-UNEXPECTED-FAIL | browser/base/content/test/performance/browser_startup_mainthreadio.js | unexpected stat on C:\Users\task_1596845536\AppData\Local\Temp\tmpudqlbh.mozrunner\wmfvpxvideo.guard before handling user events -

Comment 11

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/e77b411162c3
Ask for 3.2core+ if !REQUIRE_COMPAT_PROFILE on EGL. r=lsalzman

Comment 12

[Narcis Beleuzu [:NarcisB]]

https://hg.mozilla.org/mozilla-central/rev/e77b411162c3

Comment 13

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

Since the status are different for nightly and release, what's the status for beta?
For more information, please visit auto_nag documentation.