webgl_shadowmap_pointlight: customDistanceMaterial is no longer required

[WestLangley]

Related: #25000

What would be nice is an example where a custom distance material, or depth material, is required.

[mrdoob]

What would be nice is an example where a custom distance material, or depth material, is required.

If you can you think of and describe an example that would need a custom distance material the community can help creating it.

[WestLangley]

I am having a hard time thinking of one... We now handle the common use cases automatically. :-)

[mrdoob]

@titansoftime any ideas?

[titansoftime]

@titansoftime any ideas?

I did not think shadows on instanced meshes with materials with alphaTest would be so niche =[

If you can show me an alternate method to achieve this without customDepthMaterial, I would be more than happy to implement.

Here is how your examples showed how to do it previously:

Working r146: https://jsfiddle.net/titansoftime/hncbvgf8/

Not working r147: https://jsfiddle.net/titansoftime/q39foeLb/

This comment shows another use case from the forums:
#25000 (comment)

[WestLangley]

@titansoftime What is the reason you are you not using InstancedMesh?

[titansoftime]

I built my solution prior to the release of InstancedMesh. I have had no need to change my solution as mine works great until r147.

Are shadows using instancing with alphaTest no longer expected to work correctly with instancing unless InstanceMesh is used?

[WestLangley]

@titansoftime Hopefully we can modify r148 to accommodate your use case. But, I also expect InstancedMesh will work in your use case -- unless you are doing something I am unaware of. Would you be willing to try it?

[titansoftime]

@titansoftime Hopefully we can modify r148 to accommodate your use case. But, I also expect InstancedMesh will work in your use case -- unless you are doing something I am unaware of. Would you be willing to try it?

Yes, if you believe that is the correct way to do things in three.js, I will work on a implementing a new instancing workflow.

[titansoftime]

I have done a test with InstancedMesh. It works quite well. It is nice not having to deal with customDepthMaterial. In addition my multi-material meshes with one part alphaTest the other not (trees), now have correct shadows and respect the alpha of the foliage. Very nice.

Ideally this would be the same behavior with meshes created manually using instancedGeometry. However, InstancedMesh does simplify my workflow, so this works for me.

I wanted to also mention that this works with CSM as well which I was a little worried about.

[titansoftime]

Hmm, I am noticing a rather significant performance decrease though. I have not seen my app drop below 60fps since I got a 2070S some years ago. Now dips to low 50's. I need to investigate more what is causing this while using InstanceMesh vs. modified vertex shaders and having to use customDepthMaterial for shadows.

Edit:
TL;DR: Things are working (and now looking) correctly now I think, but a side effect is more draw calls / polygons from shadows than previous method which yielded odd shadows in some cases.

I think this performance hit is in at least part due to increased draw calls from shadows basically now casting correctly on objects that were created in more complex ways (multiple geometries merged and then using array material (one with alphaTest, the other not, then instanced).

I "believe" the customDepthMaterial method was ultimately making an incorrect shadow from a single material (foliage did not correct), and now a proper shadow is being made but now requires multiple (two in this case) draw calls for the shadow instead of one with customDepthMaterial.