radv: PLOC BVH builder
In cooperation with @bnieuwenhuizen, I present: Parallel Locally-Ordered Clustering for building BVHs!
It uses an in-shader global synchronization and task partitioning algorithm (similar to gpurt) to synchronize between iterations.
My benchmarking on a 6700XT indicates a 33% uplift for Q2RTX and the nvpro demos. Control suffers a bit from the increased build times, but still gets a tiny bit better (around 47 -> 51 FPS).