https://hgpu.org/?p=24373
Efficient Nearest-Neighbor Data Sharing in GPUs