You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The current splat renderer has a background thread that sorts all splats based on depth. This is a single cpu sort, and runs at about 5-10fps compared to the >60fps on the main render thread. For most cases, this is fine, as the user isnt moving around a scene like they are in cod, but it would still be nice to get a better algorithm. Inria uses a GPU Bitonic sort for their depth buffers, and we could try to use WebGPU to do the same, but that seems to be an issue a lot of adjacent renderers online have ran into. We could also investigate into float-based radix sorts, as most of the depths arent super large.
The current splat renderer has a background thread that sorts all splats based on depth. This is a single cpu sort, and runs at about 5-10fps compared to the >60fps on the main render thread. For most cases, this is fine, as the user isnt moving around a scene like they are in cod, but it would still be nice to get a better algorithm. Inria uses a GPU Bitonic sort for their depth buffers, and we could try to use WebGPU to do the same, but that seems to be an issue a lot of adjacent renderers online have ran into. We could also investigate into float-based radix sorts, as most of the depths arent super large.
See:
https://www.codercorner.com/RadixSortRevisited.htm
http://stereopsis.com/radix.html
https://webgpu.github.io/webgpu-samples/?sample=bitonicSort
https://github.com/hjlld/webgpu-bitonic-sort
https://gwebgpu.antv.vision/en/examples/gpgpu/basic/bitonicSort
This is pretty low priority, but it'd be cool to improve nonetheless
The text was updated successfully, but these errors were encountered: