Fix classification when terrain depth writes are enabled

[lilleyse]

The stencil buffer now stores a bit that says whether a pixel is 3D Tiles or not. This allows a classification primitive with classificationType = ClassificationType.CESIUM_3D_TILE to apply just to 3D Tiles even if the terrain depth is written. This also lets the depth plane get rendered immediately after the terrain pass which solves #6867, so 3D Tiles no longer show through the opposite side of the globe.

This is going into a staging branch and the follow up PR is going to be enabling textured poylgons and polylines on 3D Tiles, which is a lot easier with this change.

To do:

• Fix unit tests
• CHANGES.md

Next:

• Set classificationType default to BOTH everywhere.
• Optimize BOTH - Fix classification when terrain depth writes are enabled #7422 (comment)
• Optimize invert classification - Fix classification when terrain depth writes are enabled #7422 (comment)
• Textured polygons and polylines on 3D Tiles

[cesium-concierge]

Thanks for the pull request @lilleyse!

• ✔️ Signed CLA found.

Reviewers, don't forget to make sure that:

• Cesium Viewer works.
• Works in 2D/CV.
• Works (or fails gracefully) in IE11.

---

I am a bot who helps you make Cesium awesome! Contributions to my configuration are welcome.

🌍 🌎 🌏

[lilleyse]

Most of the changes in this function are just cleanup rather than bug fixes.

[bagnell]

I see 71 test failures after resetting the branch to the latest commit.

There is an optimization you could make when the classification type is both terrain and 3D Tiles without the invert classification. You could render both terrain and 3D Tiles, render the first two commands for each of the classification types, and then issue a single color draw call. Given that classification primitives are expensive to render, this might be a significant performance improvement, but it doesn't have to hold up this PR.

[lilleyse]

I'm still fixing up the unit tests and will bump when they're ready.

Yeah that idea might work. It will be a good win too because in the next PR I might set all the classificationType defaults to BOTH.

I also have ideas for speeding up invert classification. The opaque version wouldn't need a separate FBO, just one full screen pass on the main framebuffer. I think the translucent version can maybe be shrunk to a single FBO rather than two, but I haven't thought it through enough yet.

[lilleyse]

@bagnell this is ready. All tests pass locally now and the Sandcastle demos looks normal. There was a bug with translucent invert classification that is now fixed.

One change I made while writing tests was to set Vector3DTileGeometry and Vector3DTilePolygons classificationType to ClassificationType.BOTH by default. I don't think this actually affects anything other than the tests.

[bagnell]

This looks good to me. Merging this will close the bugs. Should they stay open until the staging branch is merged?

[lilleyse]

Oh good call, I'll remove those lines from the summary.

[lilleyse]

Done