https://hgpu.org/?p=16700
Memory layout in GPU implementation of lattice Boltzmann method for sparse 3D geometries