Implement PBR anisotropy per KHR_materials_anisotropy.

[pcwalton]

This commit implements support for physically-based anisotropy in Bevy's StandardMaterial, following the specification for the KHR_materials_anisotropy glTF extension.

Anisotropy (not to be confused with anisotropic filtering) is a PBR feature that allows roughness to vary along the tangent and bitangent directions of a mesh. In effect, this causes the specular light to stretch out into lines instead of a round lobe. This is useful for modeling brushed metal, hair, and similar surfaces. Support for anisotropy is a common feature in major game and graphics engines; Unity, Unreal, Godot, three.js, and Blender all support it to varying degrees.

Two new parameters have been added to StandardMaterial: anisotropy_strength and anisotropy_rotation. Anisotropy strength, which ranges from 0 to 1, represents how much the roughness differs between the tangent and the bitangent of the mesh. In effect, it controls how stretched the specular highlight is. Anisotropy rotation allows the roughness direction to differ from the tangent of the model.

In addition to these two fixed parameters, an anisotropy texture can be supplied. Such a texture should be a 3-channel RGB texture, where the red and green values specify a direction vector using the same conventions as a normal map ([0, 1] color values map to [-1, 1] vector values), and the the blue value represents the strength. This matches the format that the KHR_materials_anisotropy specification requires. Such textures should be loaded as linear and not sRGB. Note that this texture does consume one additional texture binding in the standard material shader.

The glTF loader has been updated to properly parse the KHR_materials_anisotropy extension.

A new example, anisotropy, has been added. This example loads and displays the barn lamp example from the glTF-Sample-Assets repository. Note that the textures were rather large, so I shrunk them down and converted them to a mixture of JPEG and KTX2 format, in the interests of saving space in the Bevy repository.

Changelog

Added

• Physically-based anisotropy is now available for materials, which enhances the look of surfaces such as brushed metal or hair. glTF scenes can use the new feature with the KHR_materials_anisotropy extension.

Screenshots

With anisotropy:

Without anisotropy:

[github-actions]

The generated examples/README.md is out of sync with the example metadata in Cargo.toml or the example readme template. Please run cargo run -p build-templated-pages -- update examples to update it, and commit the file change.

[torsteingrindvik]

Any particular reason to keep this as jpeg?

You likely know but: basisu -ktx2 -uastc -mipmap foo.jpeg has worked nicely for me to get mipmaps.

[pcwalton]

I think JPEG results in the smallest possible file size, and there's been a lot of discussion about keeping example files small in the repo. The original size was several MB. Mipmaps would increase size, which I didn't think was worth it as the textures look pretty good and the performance is fine.

[atlv24]

love the specular normal mixing trick, and the example is very good too. sweet beans

[atlv24]

why was this removed?

[pcwalton]

Anisotropy needs to use this function. Unlike clearcoat (the previous user of this function), anisotropy's textures aren't gated behind the pbr_multi_layer_material_textures feature, as it's not technically a separate layer. So the feature gate needs to be removed.

[IceSentry]

The volumetric_fog example crashes with this PR.

wgpu error: Validation Error

Caused by:
    In a RenderPass
      note: encoder = `<CommandBuffer-(4, 75, Vulkan)>`
    In a set_bind_group command
      note: bind group = `mesh_view_bind_group`
    Bind group 0 expects 5 dynamic offsets. However 4 dynamic offsets were provided.

[IceSentry]

The linked Filament doc uses dfg.y instead F_ab.x. I assume F_ab is the same as dfg, but is the .x vs .y correct?

[IceSentry]

Ah, I see you just copied that from somewhere else. I guess we shouldn't touch this in this PR then.

[pcwalton]

Yeah, we might want to check this later but I just copy and pasted this one.

[pcwalton]

@IceSentry volumetric_fog runs fine for me now that I've updated this PR to main. Pretty sure this was the unrelated problem with SSR from earlier.

[IceSentry]

I ran a bunch of examples again and found no new regressions

LGTM

[alice-i-cecile]

Thank you to everyone involved with the authoring or reviewing of this PR! This work is relatively important and needs release notes! Head over to bevyengine/bevy-website#1346 if you'd like to help out.