Opt-in-Per-Camera Pointlight and Spotlight Shadow Maps

[kfc35]

Objective

• Fixes Multi-Camera (Per Camera) Shadow Maps #23978
• Unblocks Fix: Support RenderLayers filtering for Lights (Adoption of #23265) #23925
• The PR description of Render point and spot light shadow maps only once, regardless of the number of cameras. #23713 mentions:

A future patch may want to add flags to cameras that specify that they should have their own point light and spot light shadow maps that inherit the render layer (and HLOD) behavior of their associated cameras. As this patch is fairly large, though, and because my immediate goal is to fix the regression in #23481, I think those flags are best implemented in a follow-up.

This is intended to be that follow-up if I understood the intention correctly 😅
This idea has also floated around on Discord, with an emphasis on this being configurable, so I preserved that desire in this PR.

Solution

A new property, has_own_shadow_maps, has been added to Camera. By default, this is set to false, preserving existing behavior.

If a camera requests its own shadow maps, new depth attachments and ShadowViews are created for these additional views in prepare_lights in light.rs, which is where the meat of this PR is. prepare_lights also manages what happens if the light itself changes or the views that opt in change. There’s also an added optimization that if all the cameras request their own shadow maps, the view-agnostic pointlight/spotlight shadow map is removed/not created.

The shader code changes are to ensure the new shadow maps are actually being used. Two new indices offset, point_spot_shadow_map_index_mult_offset and …_add_offset, are used to correctly index into the correct shadow maps. A lot of the wgsl code changes are to refactor a variable from light_id to array_index, as this more accurately describes its use and it matches the corresponding shadow sampling code for spotlight shadows. (actively soliciting for less verbose names 😆 )

Testing

For correctness:
cargo run --example testbed_3d --features=“area_light_luts” -- renderlayers
I tested with the render_layers (with the changes I put in for the light to be on the 3 render layers). It works as expected now!
I tested both a PointLight and a SpotLight (have to change the Spotlight transform to add a looking_at: Transform::from_xyz(4.0, 8.0, 4.0).looking_at(Vec3::new(3.0, 0.5, 0.0), Vec3::Y)))

I’ll probably run the bevy example runner against this PR to ensure there are no correctness regressions otherwise.

I have not done any performance testing (nor know if it’s necesssary). I don’t know if there is a performance regression in the regular “no camera has their own shadow map” case or if we should be concerned about that. If someone can help with this, I’d be indebted.

---

Showcase

The render layers scene in the 3d testbed with a PointLight now after the changes in this PR:

The same scene with a SpotLight instead described above in Testing:

[kfc35]

Hm there appears to an issue with rendering when there are multiple PointLights since my index math is messed up on the shader code side. Will figure it out

[kfc35]

I addressed the fog issues reported from the initial example runner with e9f8e87 and fixed the lighting/free_camera_controller/ multi pointlight issue with cf1ab11

So this should be ready for review finally!