CUDA capable about nanoflann HOT 2 CLOSED

This is no small task, even just for searching. If you wanted to do it, I encourage you use thrust. It would probably make more sense to fork the project, and define all stl members (e.g. std::vector) as thrust host vectors. You would then be able to relatively conveniently compute the KD tree on the CPU, and use simple assignments to thrust::device_vector's to get the built tree over there.

That said, I'd be very surprised if you got any performance gains. It will almost certainly be slower. The kinds of access patterns exhibited by a KD tree are far from ideal for GPUs. I myself have attempted this a couple times over the years, and have never gotten anything useful out of it. You're better off using an octree. It's not easy to engineer, but there's an excellent chapter on octree textures you can work off of. The use of textures makes things particularly well suited for GPUs.

from nanoflann.

I had a feeling that might be the case (after some quick googling), but thought I'd ask. I'll checkout the octree ref, thanks!

from nanoflann.