1247466 - Additional bars with transform transitions with HWA on, D2D off

Status: REOPENED

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

Description

[Oriol Brufau [:Oriol]]

Attachment: Screenshot

User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:47.0) Gecko/20100101 Firefox/47.0
Build ID: 20160210071115

Steps to reproduce:

Go to https://bugzilla.mozilla.org/attachment.cgi?id=8709506
Hover the box


Actual results:

During the transform transition, a 1px tall horizontal bar appears below the text.
I can see the bar more clearly with D3D9, but it's there with D3D11 and OpenGL too.
Sometimes I also see a 1px wide vertical bar at the left.


Expected results:

There should be no bar.

Pushlog: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=ea059733677c&tochange=2673016e7df4
Maybe bug 830347

Comment 1

[u279076]

I tested this locally on Ubuntu 15.10, Mac OS 10.11, and Windows 10; only reproducing it on Windows 10. 

> Pushlog:
> http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=ea059733677c&tochange=2673016e7df4
> Maybe bug 830347

I'm not sure how you came up with this pushlog but if it is correct, the bug would have shipped in Firefox 23. However I cannot reproduce this in Firefox 23 and only start seeing it in Firefox 37 which would seem to indicate the pushlog is not correct.

Can you please confirm the regression window is correct?

Comment 2

[u279076]

Checking this locally I came up with the following pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=035a951fc24a&tochange=f1f48ccb2d4e

This implicates the issue started in Firefox 37.0a1 on 2014-12-09. Oriol, please confirm this is the same for you.

Comment 3

[u279076]

I suspect this comes from enabling WARP assuming comment 2 is correct. Setting layers.d3d11.disable-warp to true makes this issue go away for me. Oriol, can you also confirm this?

If this is correct that implicates bug 1102499.

Comment 4

[Oriol Brufau [:Oriol]]

I use mozregression-gui on Windows 10, with these prefs:
> {
>   "browser.tabs.remote.autostart": false,
>   "browser.tabs.remote.autostart.1": false,
>   "browser.tabs.remote.autostart.2": false,
>   "layers.offmainthreadcomposition.enabled": true,
>   "devtools.chrome.enabled": true,
>   "gfx.direct2d.disabled": true,
>   "layout.css.vertical-text.enabled": true,
>   "layers.prefer-d3d9": true
> }

mozilla-central 2013-05-10 (2673016e7df4) has the problem. about:support says
> Adapter Description: Intel(R) HD Graphics 530
> Adapter Drivers: igdumdim64 igd10iumd64 igd10iumd64 igd12umd64 igdumdim32 igd10iumd32 igd10iumd32 igd12umd32
> Adapter RAM: Unknown
> Device ID: 0x1912
> DirectWrite Enabled: false (10.0.10586.0)
> Driver Date: 12-15-2015
> Driver Version: 20.19.15.4352
> GPU #2 Activ: false
> GPU Accelerated Windows: 1/1 Direct3D 11
> Vendor ID: 0x8086
> WebGL Renderer: Google Inc. -- ANGLE (Intel(R) HD Graphics 530)
> AzureCanvasBackend: skia
> AzureContentBackend: none
> AzureFallbackCanvasBackend: cairo

mozilla-central 2013-05-09 (ea059733677c) doesn't have the problem. about:support says
> Adapter Description: Intel(R) HD Graphics 530
> Adapter Drivers: igdumdim64 igd10iumd64 igd10iumd64 igd12umd64 igdumdim32 igd10iumd32 igd10iumd32 igd12umd32
> Adapter RAM: Unknown
> Device ID: 0x1912
> DirectWrite Enabled: false (10.0.10586.0)
> Driver Date: 12-15-2015
> Driver Version: 20.19.15.4352
> GPU #2 Active: false
> GPU Accelerated Windows: 1/1 Direct3D 9
> Vendor ID: 0x8086
> WebGL Renderer: Google Inc. -- ANGLE (Intel(R) HD Graphics 530)
> AzureCanvasBackend: skia
> AzureContentBackend: none
> AzureFallbackCanvasBackend: cairo

layers.d3d11.disable-warp=true does not fix the problem for me.

Comment 5

[u279076]

Okay then I guess what I'm seeing is different. Moving this back to unconfirmed.

Bas, do you have any ideas about this issue?

Comment 6

[Oriol Brufau [:Oriol]]

Also reproducible with Windows 7 and AMD Radeon HD 6290 Graphics. Same regression window.

Also reproducible with Windows XP and NVIDIA GeForce 7300 SE/7200 GS. I can't test the builds around that date because they crash directly, but the loose regresion window is consistent.

Comment 7

[u279076]

Thanks for continuing to test this, Oriol. I am now able to confirm your regression window and figured out what I was doing wrong. Firefox 23 has OMTC off by default and is needed (in addition to disabling d2d) to reproduce this bug.

To be clear, this bug only reproduces when the following prefs are set.
>   "gfx.direct2d.disabled": true
>   "layers.offmainthreadcomposition.enabled": true

I am not nominating this bug to track since it's been around since Firefox 23 and requires non-default preferences. Unfortunately I do not have inbound builds to regress this further. Bas, is it possible this is being caused by bug 830347?

Comment 8

[Bas Schouten (:bas.schouten)]

(In reply to Anthony Hughes, QA Mentor (:ashughes) from comment #7)
> Thanks for continuing to test this, Oriol. I am now able to confirm your
> regression window and figured out what I was doing wrong. Firefox 23 has
> OMTC off by default and is needed (in addition to disabling d2d) to
> reproduce this bug.
> 
> To be clear, this bug only reproduces when the following prefs are set.
> >   "gfx.direct2d.disabled": true
> >   "layers.offmainthreadcomposition.enabled": true
> 
> I am not nominating this bug to track since it's been around since Firefox
> 23 and requires non-default preferences. Unfortunately I do not have inbound
> builds to regress this further. Bas, is it possible this is being caused by
> bug 830347?

No, the D3D11 compositor only got enabled when we enabled OMTC. It's possible something in that patch messed something up but I really don't think there's value in tracking non-OMTC bugs anymore. non-OMTC isn't a supported configuration.

Comment 9

[Bas Schouten (:bas.schouten)]

(In reply to Bas Schouten (:bas.schouten) from comment #8)
> (In reply to Anthony Hughes, QA Mentor (:ashughes) from comment #7)
> > Thanks for continuing to test this, Oriol. I am now able to confirm your
> > regression window and figured out what I was doing wrong. Firefox 23 has
> > OMTC off by default and is needed (in addition to disabling d2d) to
> > reproduce this bug.
> > 
> > To be clear, this bug only reproduces when the following prefs are set.
> > >   "gfx.direct2d.disabled": true
> > >   "layers.offmainthreadcomposition.enabled": true
> > 
> > I am not nominating this bug to track since it's been around since Firefox
> > 23 and requires non-default preferences. Unfortunately I do not have inbound
> > builds to regress this further. Bas, is it possible this is being caused by
> > bug 830347?
> 
> No, the D3D11 compositor only got enabled when we enabled OMTC. It's
> possible something in that patch messed something up but I really don't
> think there's value in tracking non-OMTC bugs anymore. non-OMTC isn't a
> supported configuration.

Wait, I misread this, I thought it said main thread composition was off. So, just for the record, before bug 830347 I don't think layers.offmainthreadcomposition.enabled = true did anything on Windows. So that's probably why that appears to be the regression window.

Comment 10

[Bas Schouten (:bas.schouten)]

This looks like bug 1240800 to me.

Comment 11

[Oriol Brufau [:Oriol]]

This problem is not fixed by bug 1240800.

Comment 12

[u279076]

Oriol, does this still reproduce in the current Nightly?

Comment 13

[Oriol Brufau [:Oriol]]

(In reply to Anthony Hughes (:ashughes) [GFX][QA][Mentor] from comment #12)
> Oriol, does this still reproduce in the current Nightly?

Yes, with gfx.direct2d.disabled=true

Comment 14

[u279076]

(In reply to Oriol from comment #13)
> Yes, with gfx.direct2d.disabled=true

Thanks I am seeing this now in the latest Firefox Nightly on my Windows 10 machine. Looks like this was definitely not fixed by bug 1240800.