1817240 - Dots will appear on the box-shadow

Status: VERIFIED FIXED

(Core :: Graphics, defect)

Description

[6k64x4ma]

Attachment: test.html

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/111.0

Steps to reproduce:

• Load the attached file, then hover over the square.

Actual results:

• Dots will appear on the box-shadow.

Expected results:

• Nothing happens.

mozregression:

2023-02-17T00:17:54.687000: DEBUG : Found commit message:
Bug 1753349 (9/9): chore: update ANGLE to our fork's `firefox-111` branch r=jgilbert

Differential Revision: https://phabricator.services.mozilla.com/D162655

2023-02-17T00:17:54.693000: DEBUG : Did not find a branch, checking all integration branches
2023-02-17T00:17:54.693000: INFO : The bisection is done.
2023-02-17T00:17:54.693000: INFO : Stopped

Comment 1

[6k64x4ma]

Attachment: screenshot.png

Comment 2

[6k64x4ma]

Attachment: about_support.json

Comment 3

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

The bug has a release status flag that shows some version of Firefox is affected, thus it will be considered confirmed.

Comment 4

[Jonathan Kew [:jfkthame]]

FWIW, I tried this on my Win10 machine (ThinkStation), and was unable to reproduce it; I guess it's related to certain cards/drivers/some other configuration-specific thing.

Comment 5

[Kelsey Gilbert [:jgilbert]]

I also can't repro on Windows+AMD.
Can someone try Windows+Intel?

Comment 6

[Ashley Hale [:ahale]]

I can try on gen8 and gen11 Intel but I don't have gen9, let's see if I can repro on those at least.

Comment 7

[Glenn Watson [:gw]]

Rares, do you have HD520 hardware or similar? Are you able to reproduce this at all?

Comment 8

[Rares Doghi, Desktop QA]

Hi @Glenn, Unfortunately we dont have an Intel HD 520 but we did manage to reproduce this issue on an Intel HD 530, we also tried to reproduce this issue on our side with an Nvidia GT 730, Intel HD 630, Intel HD 2500 but it does not happen there.

On the device with Intel HD 530 we were able to reproduce this issue in Beta and Nightly only. Unfortunately we couldnt get a regression range at this time but we will try to grab one until tomorrow.

Comment 9

[Jeff Muizelaar [:jrmuizel]]

I can also reproduce this on a HD 530

Comment 10

[Rares Doghi, Desktop QA]

This is the Pushlog we got from Mozregression, Im not sure which bug is causing this issue, hope it helps.

https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=64d81842d98ded6df5d9a473453ed75f2f8e33e4&tochange=36cf8390edddbc36217a8fbe306a6d5946a5617e

Comment 11

[Kelsey Gilbert [:jgilbert]]

I can repro also:

GPU #1
Active Yes
Description Intel(R) HD Graphics P530
Vendor ID 0x8086
Device ID 0x191d
Driver Version 27.20.100.9664
Driver Date 6-1-2021
Drivers igdumdim64 igd10iumd64 igd10iumd64 igd12umd64 igdumdim32 igd10iumd32 igd10iumd32 igd12umd32
Subsys ID 06e51028
RAM 0

Comment 12

[Ashley Hale [:ahale]]

I'll be trying a few more experiments with this bug tomorrow, I'm going to try to tickle the bug (change its behavior) with several tweaks to how WebRender is binding textures, laying out blurred shadow images in the texture atlas, and other experiments to see if I can pinpoint which shader or API call is causing the breakage on this hardware. We're trying to make a decision around Tuesday next week on whether to rollback ANGLE in Beta branch, it's valuable to keep the new ANGLE running in Nightly for testing (users expect some occasional breakage in Nightly).

Comment 13

[Erich Gubler [:ErichDonGubler]]

Attachment: Bug 1817240: Revert recent ANGLE rebase (for Beta/Release only) r=jgilbert

Comment 14

[Ashley Hale [:ahale]]

Attachment: Canonicalized ANGLE D3D11 API calls on crucial frame on Intel P530 before update (firefox 110 stable)

Comment 15

[Ashley Hale [:ahale]]

Attachment: Canonicalized ANGLE D3D11 API calls on crucial frame on Intel P530 after update (firefox 112 nightly)

I took the apitrace of two versions of firefox on a critical frame where artifacts occur on Intel P530 on what I believe to be the ClearView API call, and canonicalized them (renamed addresses to make the diffs identical), here is the two apitraces if one wishes to compare the text. Not much different and certainly nothing that explains why ClearView would behave differently on a rect clear of {0, 0, 364, 364}.

Comment 16

[Ashley Hale [:ahale]]

Attachment: ClearView artifact on Intel P530 has jagged edges that may be a driver bug

Here is how the artifact looks after ClearView has been called on a render target before the red rectangle is rendered into the render target, it appears to have incorrect edge masking on the right side for this rect of {0, 0, 364, 364} which instead cleared {0, 0, 360, 364} and partially mangled {360, 0, 368, 364}.

Comment 17

[Donal Meehan [:dmeehan]]

https://hg.mozilla.org/releases/mozilla-beta/rev/acb3e383d717

Comment 18

[Rares Doghi, Desktop QA]

This issue is verified as fixed in our latest Beta Builds 111.0b7 but it still occurs in Nightly.

Comment 19

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

The bug is marked as tracked for firefox112 (nightly). However, the bug still has low severity.

:bhood, could you please increase the severity for this tracked bug? If you disagree with the tracking decision, please talk with the release managers.

For more information, please visit auto_nag documentation.

Comment 20

[Bob Hood [:bhood]]

This issue is in the recent ANGLE update. That update has been removed from Beta (Fx111), and will have a fix applied to it in Nightly (Fx112).

Comment 21

[Kelsey Gilbert [:jgilbert]]

Attachment: Bug 1817240 - Cherry-pick ANGLE skylake clearview fix.

Also fix update-angle.py:

• Use shell=True to get ninja to run
• Don't record "/PDBSourcePath:" because it depends on configuration of
  the vendoring machine and is otherwise unused, and so uselessly causes
  blame noise.

Comment 22

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/d7be17cdae33
Cherry-pick ANGLE skylake clearview fix. r=gfx-reviewers,lsalzman

Comment 23

[Cristina Horotan [:chorotan]]

https://hg.mozilla.org/mozilla-central/rev/d7be17cdae33

Comment 24

[Rares Doghi, Desktop QA]

This issue is verified as fixed in our latest Nightly 112.0a1 (2023-03-07).