1318558 - Remove D3D9 fallback

Status: RESOLVED FIXED

(Core :: Graphics: Layers, defect)

Description

[Matt Woodrow (:mattwoodrow)]

As we drop support for WinXP, we'll have basically no users on D3D9 layers.

We disabled D3D9 fallback in bug 130465, so if we reverted that (I believe the underlying issue is now fixed) we'd likely get some people using it again.

We should either get people using it again, or take the opportunity to get rid of some code.

What do you think Milan?

Comment 1

[Milan Sreckovic [:milan] (needinfo for best results)]

I'd like to re-enable it, especially if we decide to enable E10S on XP.  Need to follow up on the dependent bugs thought.  I'll leave needinfo on me to get back to it later this year, hopefully still in time for 53.

Comment 2

[Masatoshi Kimura [:emk]]

(In reply to Milan Sreckovic [:milan] from comment #1)
> I'd like to re-enable it, especially if we decide to enable E10S on XP. 
> Need to follow up on the dependent bugs thought.  I'll leave needinfo on me
> to get back to it later this year, hopefully still in time for 53.

I don't follow. Why do we enable E10S on XP while we drop support for XP? We already blocked update (bug 1317780) and will disable install (bug 1305453) on XP.

Comment 3

[Milan Sreckovic [:milan] (needinfo for best results)]

(In reply to Masatoshi Kimura [:emk] from comment #2)
> (In reply to Milan Sreckovic [:milan] from comment #1)
> ...
> 
> I don't follow. Why do we enable E10S on XP while we drop support for XP? We
> already blocked update (bug 1317780) and will disable install (bug 1305453)
> on XP.

That would be a question for Jim.

Comment 4

[Jim Mathies [:jimm]]

(In reply to Milan Sreckovic [:milan] from comment #3)
> (In reply to Masatoshi Kimura [:emk] from comment #2)
> > (In reply to Milan Sreckovic [:milan] from comment #1)
> > ...
> > 
> > I don't follow. Why do we enable E10S on XP while we drop support for XP? We
> > already blocked update (bug 1317780) and will disable install (bug 1305453)
> > on XP.
> 
> That would be a question for Jim.

We were experimenting with e10s on beta to find out what the crash raters were [1] since XP will be over on ESR for a solid year. We wanted to determine if it was safe to turn e10s on for these users there such that we would have the same features running in our normal dev branches. The crash rate regressions though for e10s+XP are pretty severe (25%-350%) so we're punting on e10s in ESR49. This said we might find ourselves in a situation where we need to apply a security fix to ESR49 that requires e10s, in which case we might still enable e10s there for XP at a future date.

[1] https://wiki.mozilla.org/Electrolysis/XP

Comment 5

[mirh]

(In reply to Matt Woodrow (:mattwoodrow) from comment #0)
> As we drop support for WinXP, we'll have basically no users on D3D9 layers.

Isn't that still a thing with people on newer Windows but older hardware?

According to Steam hardware survey 11% of users is still using a DX10 class gpu on Vista+. 
And as you might imagine this gamers-only sample is very likely skewed towards bleeding edge gpus.

Comment 6

[David Anderson [:dvander] - inactive, e-mail if emergency]

(In reply to mirh from comment #5)
> (In reply to Matt Woodrow (:mattwoodrow) from comment #0)
> > As we drop support for WinXP, we'll have basically no users on D3D9 layers.
> 
> Isn't that still a thing with people on newer Windows but older hardware?
> 
> According to Steam hardware survey 11% of users is still using a DX10 class
> gpu on Vista+. 
> And as you might imagine this gamers-only sample is very likely skewed
> towards bleeding edge gpus.

D3D11 supports D3D10 and D3D9 GPUs - we don't use the full feature set.

Comment 7

[Yuhong Bao]

FYI, Win7 still supports XP display drivers.

Comment 8

[David Anderson [:dvander] - inactive, e-mail if emergency]

TBD (to be deleted) in Firefox 55

Comment 9

[David Anderson [:dvander] - inactive, e-mail if emergency]

Attachment: part 1, rm references

Removal of obvious references to D3D9 code. The DXGID3D9TextureData class gets copied out of TextureD3D9 and into D3D9SurfaceImage.

Comment 10

[David Anderson [:dvander] - inactive, e-mail if emergency]

Attachment: part 2, rm files

To land after the merge.

Comment 11

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/7312b15a0591
Remove references to the Direct3D9 compositor. (bug 1318558 part 1, r=mattwoodrow)
https://hg.mozilla.org/integration/mozilla-inbound/rev/fbe2c62cedcb
Remove the Direct3D9 compositor. (bug 1318558 part 2, r=mattwoodrow)

Comment 12

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/7312b15a0591
https://hg.mozilla.org/mozilla-central/rev/fbe2c62cedcb

Comment 13

[Alecazam]

Just wanted to add that last I worked on DX11, "DX11 on DX9" can only access shader model 2b.  That doesn't have Vertex Texture Fetch and a lot of other features.  If your plan is to run "DX11 on DX9", you should be aware that a lot of WebGL might no longer be available to users.  You do avoid the half text offset, but the DX9 backend starts up in Shader Model 3 mode.

That being said, we're having a lot of bugs in Chrome or Angle with their DX9 backend, and you'll pave the way towards WebGL2 better.