Coriolis force implementation

Hi! I wonder is there any more optimal way to model the Coriolis force than rhis

	while (true) { // main simulation loop
		lbm.u.read_from_device();
		for (uint z = 1u; z < Nz - 1u; z++) {
			for (uint y = 1u; y < Ny-1u; y++) {
				for (uint x = 1u; x < Nx - 1u; x++) {
					const uint n = x + (y + z * Ny) * Nx;
					lbm.F.x[n] = -Co * lbm.u.y[n];
					lbm.F.y[n] =  Co * lbm.u.x[n];
					lbm.F.z[n] =  0.0f;
				}
			}
		}
		lbm.F.write_to_device();
		lbm.run(1u);
	}

It slows down a thousand times :(

Start from the simulation loop shown in the issue and trace lbm.u.read_from_device(), lbm.F.write_to_device(), and lbm.run(1u). Measure the individual transfer and loop costs, then compare possible implementation approaches while preserving the Coriolis force behavior. Done means the force is modeled correctly without the reported severe slowdown, supported by a reproducible benchmark.