1714897 - Fix glxtest EGL code path on amdgpu

Status: VERIFIED FIXED

(Core :: Graphics: WebRender, defect)

Comment 1

[Glenn Watson [:gw]]

Attachment: Bug 1714897 - Fix glxtest EGL code path on amdgpu

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/cbbccab13f24
Fix glxtest EGL code path on amdgpu r=rmader

Comment 3

[Noemi Erli[:noemi_erli]]

https://hg.mozilla.org/mozilla-central/rev/cbbccab13f24

Comment 4

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

Set release status flags based on info from the regressing bug 1714069

Comment 6

[Julien Cristau [:jcristau] (back April 22)]

There's an uplift request for the regressing bug, so presumably there should be one here too.

Comment 7

[Pascal Chevrel:pascalc (PTO until April 26)]

Glenn, do we need to uplift this fix to beta and release as we is a request to uplift bug 1714069 which caused this regression?

Comment 8

[Glenn Watson [:gw]]

Comment on attachment 9225499 [details]
Bug 1714897 - Fix glxtest EGL code path on amdgpu

Beta/Release Uplift Approval Request

• User impact if declined: Linux users on AMD GPUs get software webrender instead of hardware webrender.
• Is this code covered by automated tests?: No
• Has the fix been verified in Nightly?: No
• Needs manual test from QE?: Yes
• If yes, steps to reproduce: Per bug report (check enabled compositor in about:support).
• List of other uplifts needed: Bug 1715245
• Risk to taking this patch: Medium
• Why is the change risky/not risky? (and alternatives if risky): The change itself is small, self contained and simple. The risk is that this code is sensitive to platform differences such as GPU, driver version etc and we don't have great test coverage of all those different hardware permutations. It can't affect users on Windows / Mac though, which should reduce the risk.
• String changes made/needed:

Comment 9

[Rares Doghi, Desktop QA]

Hi Glen, just to be clear for Linux users with AMD GPU in about support its suppose to say

WEBRENDER_SOFTWARE- available by default ? instead of disabled ?
or
Compositing - Webrender software instead of just Webrender ?
WEBRENDER_COMPOSITOR - available ? instead of disabled ?

the reason we are asking is because on an Ubuntu 18.04 machine with AMD RS780 it just says disabled and I think this GPU might not support webrender.

We also tried on an ubuntu 20.04 with AMD rx 550 and we couldnt see any differences between older builds and the later ones, Is this specific to certain AMD cards ?

I will attach the About:support files for both machines and maybe you can take a look.

Comment 10

[Rares Doghi, Desktop QA]

Attachment: RS780Support.txt

Comment 11

[Rares Doghi, Desktop QA]

Attachment: RX550Support.txt

Comment 12

[Julien Cristau [:jcristau] (back April 22)]

Comment on attachment 9225499 [details]
Bug 1714897 - Fix glxtest EGL code path on amdgpu

approved for 90.0b7

Comment 13

[Julien Cristau [:jcristau] (back April 22)]

https://hg.mozilla.org/releases/mozilla-beta/rev/9774b3e5a0bd

Comment 14

[Glenn Watson [:gw]]

Hi Rares, the main thing we're fixing / checking for here is that we don't end up selecting Software Webrender as compositing backend on recent AMD cards.

What was causing this was when glxtest would return mesa/llvmpipe as selected driver for the 'Driver Vendor' string instead of mesa/radeonsi.

It's going to be a bit difficult to confirm this, since it's dependent on GPU model (not sure exactly which ones are affected), and possibly mesa version etc.

From what I can tell, this looks correct in the support attachments above.

Comment 15

[Rares Doghi, Desktop QA]

Hi, thank you for answering, do you by any chance have that AMD GPU and see if the mesa/radeonsi is now displayed correctly on your end for our latest Nightly and Beta builds? we do not have the right AMD card for this issue and we cannot confirm the fix on our end.

Comment 16

[Glenn Watson [:gw]]

Yes - I have an affected GPU - I can confirm it now looks correct for me.

Comment 17

[Rares Doghi, Desktop QA]

Thank you for checking this, I will Update the flags based on Comment 16 , should we also set 89 to affected ?

Comment 18

[Glenn Watson [:gw]]

Yes, I believe 89 is also affected. Thanks!

Comment 19

[Rares Doghi, Desktop QA]

Marking 89 as affected as well based on Comment 18. Thank you Glenn, Sorry to bother you with all this, In case this gets uplifted we might need you to check this fix again on the affected GPU, I apologize in advance.

Comment 20

[Robert Mader [:rmader]]

(In reply to Rares Doghi from comment #19)

Marking 89 as affected as well based on Comment 18. Thank you Glenn, Sorry to bother you with all this, In case this gets uplifted we might need you to check this fix again on the affected GPU, I apologize in advance.

Actually 89 is not really affected as the regressing bug 1714069 was fixed differently for 89 (bug 1715902). So if this bug, 1714897, gets backported as well, then we'd also need bug 1715245 - and I think the decision was to not do that (hence bug 1715902).

Comment 21

[Rares Doghi, Desktop QA]

Thank you Robert, I will update the remaining flags.