Document how to configure cascaded shadow maps in different situations

[superdump]

In the cascaded shadow maps #7064 PR, there were some situations raised that require or would benefit from different cascade configuration to the default which focuses on a perspective camera at head height.

One example was a top-down or isometric type view where a lot of the scene objects are within a small depth range and distributed across the screen with no objects close to the camera. In this case probably fewer or even just one cascade with a higher shadow map resolution could be a better choice.

I think it would be good to try out such situations, tweak the settings, maybe adjust the projection’s near plane as @robtfm suggested in the PR discussion as that impacts the view frustum and if I remember correctly also the cascade projection bounds.

[rparrett]

Here's a little playground I was using: https://gist.github.com/rparrett/0b8a2de7443d41941ad6463e2a42c331

I'm not sure if the cascades are following the near plane. edit: they definitely don't

[danchia]

@rparrett you're right - it currently doesn't follow the near plane, but that shouldn't be too large a change. Just need to fix this

bevy/crates/bevy_pbr/src/light.rs

Line 378 in c9a53bf

0.0

line most likely.

@superdump if it feels more natural, I think it would make sense to allow users to configure the near plane direct in CascadesShadowConfig

[rparrett]

Just need to fix this

I attempted that (I think the other bounds need the near plane added as well), but I think there must be something more to it, because the shadows were not even wiggling when adjusting the near plane in the playground above.

At the time I believe this might be due to the "raw bounds" being used here:

bevy/crates/bevy_pbr/src/render/light.rs

Line 1231 in c9a53bf

.zip(&light.cascade_shadow_config.bounds)

But I was in a bit over my head.

Additionally, there's a question over what shadow_maximum_distance should mean. If it's meant to be "distance from the start of the first cascade" then I think calculate_cascade_bounds needs to be aware of the near plane. Either way, the docs could use a bit of clarification.

[superdump]

shadow_maximum_distance is view z distance from the camera position

[superdump]

@danchia that sounds reasonable and you’re probably right that that line you referenced plays a significant role in the calculation of the cascade projections.

[superdump]

Though perhaps if adjusting that near plane, we’d need to adjust the way the cascades are calculated in general. The first cascade far bound needs to be > the near plane, and then subsequent cascades need to have far bounds coming after that that scale in a sensible way across the range.