gpu sorting (radix)

[taher]

what's stopping to port sorting on the gpu? reference: aras-p/UnityGaussianSplatting#1

[Kevin Kwok]

I've seen implementations of radix sort for WebGPU that work well, but that's not well supported by different browsers— this project sticks with WebGL 2.0 for broad compatibility.

For the experiments I've done with radix sort on webgl, it's comparable to the current webworker implementation and wouldn't lead to a performance improvement.

There are a few additional sorting improvements that are possible e.g. with WebAssembly, such as the one that's used in https://github.com/huggingface/gsplat.js