WebGPURenderer: Add PCFShadowMap support.

[Mugen87]

Related issue: -

Description

This PR makes sure PCFShadowMap can be used with WebGPURenderer. It is also the default shadow map type now similar to WebGLRenderer (previously the default type was null).

I have slightly refactored AnalyticLightNode to make the code more consistent to what we had before in shadowmap_pars_fragment.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

Filesize dev	Filesize PR	Diff
684 kB (169.4 kB)	684 kB (169.4 kB)	+0 B

🌳 Bundle size after tree-shaking

Minimal build including a renderer, camera, empty scene, and dependencies.

Filesize dev	Filesize PR	Diff
460.9 kB (111.2 kB)	460.9 kB (111.2 kB)	+0 B

[Mugen87]

I'm not sure about the root cause yet but lights with WebGPURenderer seems to need a shadow bias more often. I suspect this is related to the shadow compare operation which is different compared to WebGLRenderer. The previous approach used MeshDepthMaterial with RGBA packing, the new approach uses a depth texture attachment.

Changing the depth texture type to THREE.FloatType did not help.

[Mugen87]

Side note: I have observed a few issue with shadow casting lights and WebGPURenderer that we might want to solve with different PRs.

1. It is currently not possible to dynamically toggle mesh.receiveShadow.
2. Dynamically changing light.castShadow is not supported. It should be as follows:
   • When castShadow is set to false for an already shadow casting light, the shadow should be completely removed. Right now, the shadow map is just not updated anymore but objects are still affected by the shadow.
   • When a light does not cast shadows but then castShadow is set to true, the shadow map should be rendered like with an initial setup. Right now, the shadow is never rendered.

[Mugen87]

It is currently not possible to dynamically toggle mesh.receiveShadow.

I've tried to fix this via be1bd9f which is inspired by WebGLRenderer. An object uniform is now used to control whether an object should receive shadows or not. But since shadowNode is cached, other objects now have a wrong shader. shadowNode would have to be maintained per object and not per light.

I'll revert the commit for now and save this task for another PR. My hope was it would be an easy fix but it seems more thought is required to get this work.

[sunag]

At the time I wrote this code I used mix instead of conditional because we hadn't added diagnostic( off, derivative_uniformity );, I still need to know more about possible side effects of this, but automatic stack conditionals like If are not compatible with caching in variables, since it is not possible to record the process flow as we would do in a chain.

There are more items for me to review just describing in case I forget.

[sunag]

It's possible now we have custom filters per light.

spotLight.shadow.filterNode = tslFn( ( { depthTexture, shadowCoord } ) => {

	return texture( depthTexture, shadowCoord.xy ).compare( shadowCoord.z );

} );

PCF and Basic shadow map