Optimization of Cache Oblivious Lattice Boltzmann Method in 2D and 3D

The Lattice Boltzmann Method (LBM) is one of the modern techniques to simulate fluids. A lot of work has already been done to speed up the LBM on cache-based microprocessor architecture. In order to achieve high performance on these architectures, not only the clock rate of the processor plays an important role, but also the significant influence of the caches have to be taken into account. A common method to exploit spatial and temporal data locality by reusing the cache is blocking (a cache aware algorithm). In this paper, a “ Cache Oblivious Lattice Boltzmann Algorithm (COLBA) ” and its optimization techniques are presented. Therefore Frigo’s stencil based cache oblivious algorithm [FS05] has been coupled to the LBM. Main aim of this work has been the comparison of cache aware/blocked LBM and COLBA on different architectures. The results of the COLBA have shown equivalent performance to the fastest LBM implementations available at the Regional Computing Center at Erlangen (RRZE) [Don04] also less cache misses and less buss access than the iterative LBM implementation.