PMREMNode: Manage own generator, fix caching.

[Mugen87]

Fixed #30503.

Description

Managing a global instance of PMREMGenerator in the renderer solves issues when more than one renderer is used in a multi-canvas setup.

I have updated the PR since integrating PMREMGenerator in the renderer is bad for tree-shaking. Since there are usually not many PMREMNode instances per scene it should be safe when each node manages its own generator.

[github-actions]

📦 Bundle size

Full ESM build, minified and gzipped.

	Before	After	Diff
WebGL	336.42 78.33	336.42 78.33	+0 B +0 B
WebGPU	518 143.91	518.01 143.8	+12 B -110 B
WebGPU Nodes	517.46 143.8	517.48 143.69	+12 B -111 B

🌳 Bundle size after tree-shaking

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

	Before	After	Diff
WebGL	465.41 112.17	465.41 112.17	+0 B +0 B
WebGPU	589.87 160	589.88 159.76	+11 B -240 B
WebGPU Nodes	545.24 149.54	545.25 149.23	+11 B -314 B

[x0YYYY0]

Thank you 💯 !

[Mugen87]

@sunag This check causes issues in the existing implementation (so without this PR) since depending on the scene setup the reflections are incorrectly flipped along the x-axis. Here is a slightly modified version of webgpu_cubemap_dynamic that demonstrates the issue:

https://jsfiddle.net/L7zn6bqs/1/

I understand you have added the check in #27988 to fix webgpu_cubemap_dynamic in the first place but it seems a different approach is needed. When each PMREMNode has its own generator, the check should no longer be required.

[Mugen87]

The PR also moves the texture cache in the class scope since otherwise a similar error like #30503 can occur when the same instance of Texture is used in two renderers. One renderer would generate a PMREM texture that would be shared by another one which does not work. PMREM textures are render target textures and WebGL/WebGPU resources can't be shared across contexts.

[sunag]

The PR also moves the texture cache in the class scope since otherwise a similar error like #30503 can occur when the same instance of Texture is used in two renderers.

I think it's better to relate the cache to the renderer, for example cache[ renderer ][ texture ], if we leave it in class scope we will probably have duplicate PMREM.

[Mugen87]

Great idea! I've added the cache back which is now hierarchical and respects the current active renderer.