1258082 - Right hand side of youtube video just lines

Status: RESOLVED INCOMPLETE

(Core :: Audio/Video: Playback, defect, P5)

Description

[Brian Birtles (:birtles)]

Attachment: youtube-1.png

I started noticing this since yesterday on Youtube videos. It doesn't always happen, but once it starts happening it happens for all Youtube videos even ones embedded in other sites.

See attached screenshot.

The only common factor I can think of is that in each case I had Netflix open in another tab (but not actually playing a video).

Comment 1

[Brian Birtles (:birtles)]

Attachment: youtube-2.png

Screenshot including video stats.

Comment 2

[Brian Birtles (:birtles)]

Attachment: about:support

Comment 3

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Matt - is this a D3D11 DXVA issue?

Comment 4

[Brian Birtles (:birtles)]

Actually, I think I might have seen this show up in Chrome on a different video later. So this might not be Firefox-related after all.

Comment 5

[Matt Woodrow (:mattwoodrow)]

Chrome generally uses WebM on youtube, so it seems unlikely that they'd have the same issue.

It could be d3d11 related, but that's been enabled for a few weeks now.

Does setting media.windows-media-foundation.allow-d3d11-dxva=false stop the problem from happening?

The images are weird. It looks like we're clamping the texture coordinate, but we still get some colours showing up in the clamped area.

I guess incorrect clamping in the YUV->RGB conversion might have that effect.

Comment 6

[Brian Birtles (:birtles)]

(In reply to Matt Woodrow (:mattwoodrow) from comment #5)
> Does setting media.windows-media-foundation.allow-d3d11-dxva=false stop the
> problem from happening?

I'll give it a try tonight. This only reproduces on my home machine (which is running Aurora, by the way).

Comment 7

[Brian Birtles (:birtles)]

(In reply to Brian Birtles (:birtles, away 2-10 April) from comment #6)
> (In reply to Matt Woodrow (:mattwoodrow) from comment #5)
> > Does setting media.windows-media-foundation.allow-d3d11-dxva=false stop the
> > problem from happening?

Yes, that fixes the problem.

Comment 8

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Chris - can you make sure his driver is blacklisted?

Comment 9

[Matt Woodrow (:mattwoodrow)]

As I said on irc, it's hard to know what exactly to blacklist here.

If it's a hardware problem, then we'll likely need to blacklist the entirety of the chipset generation, not just this exact device id.

If it's a driver bug, then we want to know the range of driver versions affected by it, it's very unlikely to be just this exact one.

Blacklisting just this exact configuration will likely miss the majority of users affected in the wild, while hiding it from our developer that can reproduce it.

Brian, would you be willing to try looking for a new driver on the intel website and seeing if that helps? Yours isn't particularly old, but it's believable that a newer one has been released.

Given that we only have a single report, I wouldn't assume that this is a huge problem.

If we have reason to think it is common, then we really need a way to gather more information about which system it affects. Extending the startup sanity test to check for this pattern (might need a new test video) and reporting telemetry for failures might be the best option since our current QA approaches aren't catching it.

We'd also probably want a D3D11 DXVA specific blacklist, since falling back to d3d9 works fine for Brian.

Comment 10

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Matt - would it be worth trying a sanity test to detect this problem?

Comment 11

[Matt Woodrow (:mattwoodrow)]

(In reply to Anthony Jones (:kentuckyfriedtakahe, :k17e) from comment #10)
> Matt - would it be worth trying a sanity test to detect this problem?

Comment 1 suggests that this generally doesn't happen initially, so a sanity test to catch it would be difficult.

Without more reports of this happening to other people I doubt it's worth the engineering effort.

Comment 12

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Joe - which driver versions or hardware combinations should we be using D3D9 DXVA2 in preference over D3D11 DXVA2?