Depth testing across layers

[avaer]

The explainer mentions depth testing is not to be performed across layers. What is the reasoning behind that? Is it because the platforms targeted currently do not support depth buffers for projection layers?

Also, could it be allowed as an optional feature now or in the future?

[cabanier]

The explainer mentions depth testing is not to be performed across layers. What is the reasoning behind that? Is it because the platforms targeted currently do not support depth buffers for projection layers?

The depth buffers are only supported to do reprojection. They aren't used to "sort" pixels between layers.

Also, could it be allowed as an optional feature now or in the future?

Do you know of any runtime that allow this? Would you expect this sorting to also take the other layers into account?

[avaer]

Yes, the Exokit runtime utilizes WebXR layers for per pixel depth sorting, as a traditional depth test. It does depend on the order of layers due to transparency.

As far as I could tell the OpenXR depth extension does not preclude this usage, and folks at PlutoVR and Monado are experimenting with it as a feature (I could be wrong on this point, I'll ask them to chime in on this issue).

It would be good if WebXR was in alignment either way.

[cabanier]

As far as I could tell the OpenXR depth extension does not preclude this usage, and folks at PlutoVR and Monado are experimenting with it as a feature (I could be wrong on this point, I'll ask them to chime in on this issue).

Yes, it would be good to know if they are working on a hardware compositor that supports this and if so, how it works

It would be good if WebXR was in alignment either way.

I agree.

[cabanier]

As far as I could tell the OpenXR depth extension does not preclude this usage

hmm, actually it does:

The XR runtime may use this information to perform more accurate reprojections taking depth into account.

It doesn't say anything about depth sorting between layers

[avaer]

Yes, it would be good to know if they are working on a hardware compositor that supports this and if so, how it works

I don't think it's a hardware compositor, but driver software.

It doesn't say anything about depth sorting between layers

It doesn't specify the manner of the reprojection, nor does it SHOULD NOT any particular mode so it is definitively not precluding (making impossible) anything. Arguably, sorting the layers provides the most accurate reprojection of the layer, so using it for sorting falls into both text and spirit of the MAY.

[bradgrantham-lunarg]

Reprojection is the warping of a single composition layer to better match the viewer's head orientation at the time of scanout. Putting together multiple layers into an image is composition. So if the spec says The XR runtime may use this information to perform more accurate reprojections taking depth into account, it means that the information may or may not be used to warp the single layer better.

I think there's probably a place for a "depth-aware" composition extension.

[cabanier]

@avaer I will close this issue for now. Please let me know if there's a UA that wants to implement this.

[avaer]

We mentioned Exokit already, I'm not sure why the main usage of WebXR layers is ignored.

I'll ping the others again.

[cabanier]

@avaer Can you create a PR and put it on the agenda so we can discuss?

[avaer]

Yes, I can do that, though I have a conflict for tomorrow's teleconference.

Did you want to move it to next week or still discuss it tomorrow?

[cabanier]

If you can't make it, let's move it to the next CG call in 2 weeks (unless @AdaRoseCannon can cover it in the WG call next week).
Once you create the PR, we can begin the discussion here.