Repository navigation
Optional gpulite device path: linkcell::gpu::Workspace::knearest_into writes the same packed n * k indices as the host walk. Occupancy defaults to threads-per-particle 8 (HOOMD / vesin), overridable by LINKCELL_TPP and LINKCELL_BLOCK. Setup copies ride the workspace stream. Orthorhombic cells, k <= 16.