Materialx upgrade

[bhouston]

Description

fixes #20541

The goal is to increase the parity of the MaterialX implementation. I guided this improvement via this Fidelity Test website that compared, using +300 MAterialX samples, the reference MaterialX Viewer compared to the current Three.JS MaterialXLoader and my improved MaterialXLoader.

https://material-fidelity.ben3d.ca

You can see that the new loader has basically perfect MaterialX reproducibility for pretty much everything:

The number of bugs addressed is quite massive:

• Most noise implementations were incorrect compared to reference, so they have all been improved.
• Dozens of minor fixes to the mx_* functions, and updating their defaults to match MaterialX expectations.
• The conditions, mx_ifgreater/mx_ifgreatereq/mx_ifequal, were giving the opposite intended result.
• Rotate 2d, rotate 3d, place 2d were all reimplemented to match reference.

It also introduces these new features:

• A modular design with a separate node registry.
• New surfaces open_pbr_surface, and gltf_pbr in addition to standard_surface.
• Archive support, the ability to load .mtlx.zip files, such as those from AMD Material Library and PolyHaven directly.
• Generates a report if there are warnings/errors.
• Properly handling of default values.
• Add support for hextile and hextilenormapmap.

Updated MaterialX noise example with all noise types showcased:

This contribution is funded by Land of Assets

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	365.32 86.78	365.32 86.78	+0 B +0 B
WebGPU	656.3 181.39	659.93 182.55	+3.63 kB +1.16 kB
WebGPU Nodes	654.42 181.11	658.05 182.27	+3.63 kB +1.16 kB

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	498.04 121.5	498.04 121.5	+0 B +0 B
WebGPU	728.29 196.33	728.29 196.33	+0 B +0 B
WebGPU Nodes	677.5 183.73	677.5 183.73	+0 B +0 B

[bhouston]

This is ready for review: @sunag. I just added textile support.

[bhouston]

FYI: I've just identified a bug in the reference renderer MaterialXView with regards to some UV issues: AcademySoftwareFoundation/MaterialX#2881. This bug has likely leaked into ThreeJS because I was trying to match those results exactly. I will try to patch this in the next day or so.

[bhouston]

Okay, I've completed my investigation of the UV space differences between MaterialX, Blender, ThreeJS. It turns out that ThreeJS allows both "bottom-left" and "top-left" UV spaces depending on how you load the assets.

If you load glTF assets (meshes/textures), you get "top-left" which is the glTF native uV space, but if you create ThreeJS native assets and materials outside of the glTF loader, you get "bottom-left" uv space.

Full details here: #33531

What this means is I will support both UV spaces in the MaterialXLoader so it can work on glTF and native ThreeJS assets.

[mrdoob]

Amazing work here Ben!

[sunag]

Some materials like standard_surface_color3_vec3_cm, standard_surface_brick_procedural and some others they are no longer working as expected in webgpu_loader_materialx, unlike your test link.

Could there be some difference in the code or material sample here?

webgpu_loader_materialx

https://material-fidelity.ben3d.ca/