Krzysztof Narkowicz

GPGPU question! in my modeling tool, I need to add a way to box select vertices, *without* selecting those obscured behind other geometry my current solution is a brute force CPU side intersection test of the selection box, and then an intersection test for every single face, for every single vertex, which isn't very scalable I could get into multithreading it on the CPU and burst compile and create acceleration kd-trees and uuuh [unhappy freya noises] or maybe, I can do this with compute shaders? 1. render a depth texture of the scene 2. run a compute shader that depth tests all vertices 3. the compute shader then writes a single bit for its selection state into a buffer 4. download to the CPU this would *significantly* speed up computation, but now the new worry is whether downloading the results of that will be fast enough for 8192 vertices (unusually high for my tool), this would be 1 KiB of data - would that take a reasonably short amount of time to download? I guess I could do things like run-length encode the thing, or other types of compression, which would reduce the average amount of data needed by quite a lot, but I don't know how to tell the CPU how much to download, before the download, and so maybe that would just incur two delays instead of just one, so idk if that's worth it