Globe covering tiles optimization

[kubapelc]

Improves the performance of globe's coveringTiles function, adds benchmarks to measure globe coveringTiles performance, also likely slightly improves mercator's coveringTiles performance (not measured, but some of the changes should help there too).

Improvements:

• Adds a cache for tile AABBs to avoid recomputing them every time. The cache is cleared at the beginning of every frame, only keeping the tile AABBs that were accessed in the last frame. Any two consecutive frames should mostly be viewing the same set of tiles, so most of the cache should be reused. (Only the AABB cache idea from Globe: symbol and coveringTiles optimizations #4778 is implemented, not the accurate tile-frustum intersections.)
• Improves performance of plane-AABB intersection test
• Improves frustum-AABB performance by applying the pervious optimization to the frustum's 6 planes
• Simplifies the final test in frustum-AABB intersection by precomputing the frustum's bounding box

Performance

There are two benchmarks, one with the camera looking at the globe from above, one with a pitched camera. The camera slowly pans to the east during the benchmark. Here are the rough contributions of individual changes:

Change	Unpitched	Pitched
Baseline	13.5 ms	21.0 ms
Cached AABBs	9.7 ms	14.0 ms
+FastAABBplane	8.8 ms	12.2 ms
+FastFrustum	7.4 ms	9.8 ms
+FrustumBounds	7.4 ms	9.5 ms

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 89.08%. Comparing base (c8a5ea8) to head (35fe7fc).
Report is 9 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4937      +/-   ##
==========================================
- Coverage   89.13%   89.08%   -0.06%     
==========================================
  Files         267      269       +2     
  Lines       38234    38268      +34     
  Branches     2347     2356       +9     
==========================================
+ Hits        34081    34090       +9     
- Misses       3151     3174      +23     
- Partials     1002     1004       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[HarelM]

The following PR is nearing its merge:

• Fix level of detail at high pitch #4779

If you need, I can create a branch in this repo for collaboration, so you could open this PR against the updates made in the above PR, but otherwise, I prefer to have the above PR merged first to avoid conflicts on the hard work @NathanMOlson did.

[kubapelc]

Lets wait for #4779 to be merged first.

[birkskyum]

@kubapelc , does this affect tile loading of the globe when the terrain3d is active too?

[kubapelc]

does this affect tile loading of the globe when the terrain3d is active too?

It should not, this PR does not change functionality of tile culling, just makes it faster.

But globe coveringTiles so far assumes that there is no terrain and does not take tile min/max elevation into account when generating AABBs. I think it should not be an issue for the purposes of porting terrain to globe.

I'd like to redo globe coveringTiles to not use AABBs, since they do not approximate tiles on a sphere very well, and use oriented bounding boxes instead. I would add support for tile min/max elevation along with this.

[birkskyum]

@kubapelc , sounds great - note the terrain is actually fully working at this point on globe, except for an issue with the mesh near the poles #4984

• Support Terrain in Globe projection #4977

[HarelM]

Converting tiles have been merged, this can now proceed.

[HarelM]

Can you add some tests and a changelog entry?

[kubapelc]

Can you add some tests and a changelog entry?

I've added both. Most of the changes were already covered by existing tests, I've added tests for aabb-plane intersection and for the aabb cache.

[HarelM]

This looks straight forward and good, thanks!
Added a few minor comments.