1962753 - <https://ulucode.com/random/webgputests/linked>: After a few seconds, frame drawing stops. Several frames are duplicated and no further progress is made.

Status: VERIFIED FIXED

(Core :: Graphics: WebGPU, defect)

Description

[Alice0775 White]

Steps to reproduce:

1. Navigate on https://ulucode.com/random/webgputests/linked/
2. Observe the page

Actual Results:
After a few seconds, frame drawing stops. Several frames are duplicated and no further progress is made.
See screencast: https://youtu.be/7BhgRq7M1aw

Regression window:
https://hg-edge.mozilla.org/integration/autoland/pushloghtml?fromchange=d795f74ad10b1802f27c50f547735b7c9fa676a6&tochange=93c4b2524bf66523d9b2427b7a058d9d3e5fd365

Comment 1

[Alice0775 White]

Attachment: about:support

Comment 2

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

:jimb, since you are the author of the regressor, bug 1962269, could you take a look?

For more information, please visit BugBot documentation.

Comment 3

[Erich Gubler [:ErichDonGubler] (he/him)]

Changing NI to :teoxoy, who authored our recent OOM handling machinery.

Comment 4

[Teodor Tanasoaia [:teoxoy]]

Attachment: before.png

The regression window contains https://github.com/gfx-rs/wgpu/pull/7472 but even with the memory pressure detection added by that PR disabled, the behavior of the app is unchanged.

I went to d795f74ad10b1802f27c50f547735b7c9fa676a6 (the beginning of the regression window) and I can reproduce the errors in the console (see attachment). There is one major difference: the app doesn't freeze - that's because https://github.com/gfx-rs/wgpu/pull/7472 will cause the device to become lost on out of memory errors caused outside of allowed methods (ex: createBuffer/createTexture).

Comment 5

[Teodor Tanasoaia [:teoxoy]]

I will investigate where the initial OOM comes from.

Comment 6

[Bogdan Maris, Desktop Test Engineering [OOO until 14th April]]]

I reproduced the issue described in comment 0 using an old Nightly build from 2025-04-25. I used the latest Nightly 141.0a1 and Firefox 140.0b1 builds on the same Windows 11 machine where I reproduced the issue and I can still see that the frame freezes for 1-2 seconds from time to time on both builds. I made a profiler recording on Nightly for it maybe it helps: https://share.firefox.dev/45rh6PG

I am using a AMD Radeon RX 7900 XTX GPU with a Radeon 7 7800x3d CPU.

I also noticed that in 140.0b1 we don't paint anything in the page with the cursor when hovering over the hexagons from the page, only the hexagons are glowing. Here is the console from both Beta 140.0b1 and Nightly 141.0a1.
Should we call this bug as verified fixed based on the above or will this be further investigated here? Needinfo on :teoxoy since there is no assignee on this bug.

Beta 140.0b1

WARNING: Multiple instances of Three.js being imported. index-kjH2I029.js:482:58640
 Renderer : 
 Object { isRenderer: true, domElement: canvas#threecanvas, backend: {…}, samples: 4, autoClear: true, autoClearColor: true, autoClearDepth: true, autoClearStencil: true, alpha: true, logarithmicDepthBuffer: false, … }
index-kjH2I029.js:491:14241
Uncaptured WebGPU error: Internal error: FXC D3DCompile error (Unspecified error (0x80004005)): F:\Desktop\NEWWWWWW\eeeeeeeeeeeeeeeeeeee\firefox\Shader@0x00000184E4367000(17,12-33): error X3059: array dimension must be between 1 and 65536
Uncaptured WebGPU error: In a set_pipeline command, caused by: ComputePipeline with '' label is invalid
Uncaptured WebGPU error: Command encoder is locked by a previously created render/compute pass. Before recording any new commands, the pass must be ended.
Uncaptured WebGPU error: CommandBuffer with '' label is invalid

Nightly 141.0a1:

WARNING: Multiple instances of Three.js being imported. index-kjH2I029.js:482:58640
Renderer : 
Object { isRenderer: true, domElement: canvas#threecanvas, backend: {…}, samples: 4, autoClear: true, autoClearColor: true, autoClearDepth: true, autoClearStencil: true, alpha: true, logarithmicDepthBuffer: false, … }
index-kjH2I029.js:491:14241

Comment 7

[Teodor Tanasoaia [:teoxoy]]

It seems we don't use DXC on Beta, https://phabricator.services.mozilla.com/D251591 will fix that.

As for the occasional stuttering it looks like there is a lot of GC activity that is blocking rendering, but this needs more investigation.

Comment 8

[Bogdan Maris, Desktop Test Engineering [OOO until 14th April]]]

Let me know if you want me to log a new issue with the occasional stuttering.

Comment 9

[Teodor Tanasoaia [:teoxoy]]

A new issue would be great since it's a different issue than the initial one reported in this bug. We can then close this one since Bug 1926651 also landed. Thanks!

Comment 10

[Bogdan Maris, Desktop Test Engineering [OOO until 14th April]]]

(In reply to Teodor Tanasoaia [:teoxoy] from comment #9)

A new issue would be great since it's a different issue than the initial one reported in this bug. We can then close this one since Bug 1926651 also landed. Thanks!

Logged bug 1969834 and cc'd you in it. Closing this bug based on the above comments.