A Spatial Lattice Simulation Kernel

The Problem: A new technique for mapping spatial lattice data into a hierarchical memory system has been devel-oped. This technique makes possible high-performance lattice-computation hardware. Can the same technique also be used as the basis of efficient software simulators? Motivation: Many interesting computations operate on uniform spatial lattices, or on a coupled hierarchy of such lat-tices. These include finite difference calculations, image processing and rendering algorithms, functional and physical logic simulation, and cellular automata simulations of a wide variety of physical systems and virtual-reality systems (see Figure 1). The kinds of processing used in these simulations includes both numeric and symbolic operations performed in a uniform manner on regularly spaced sets of lattice sites[2, 3]. Figure 1: Left: A lattice-gas hydrodynamic simulation. Discrete particles conserve momentum as they flow past a half-cylinder. A smoke-fluid (white) is also simulated, to visualize the flow. Right: A 3D bit map is rotated on a lattice using discrete shift operations, and rendered using discrete lattice light-particles. Previous Work: A recently proposed spatial lattice architecture[3] introduces a new technique for mapping lattice data into a memory hierarchy with a granular structure (words, rows, etc.). This technique enables large and efficient lattice simulations on specialized hardware. This technique also makes it possible to avoid much of the overhead that is normally seen in performing non-numerical lattice computations on conventional computers