论文标题
减少流体动力无网格方法中冗余的相互作用的数量
Reducing the number of redundant pair-wise interactions in hydrodynamic meshless methods
论文作者
论文摘要
广泛使用的拉格朗日数值代码，这些代码计算与相邻分辨率元素（颗粒）的物理相互作用，重复了对粒子对之间相互作用的计算。我们开发了一种减少冗余计算数量的算法。该算法利用哈希功能来标记已经计算的交互和最终碰撞。哈希的结果存储在两个缓存中。不用限制缓存内存的使用情况，就可以避免所有重复的计算，从而实现了两个因素的加速。我们表明，将缓存大小（以位）限制为典型的相邻粒子数量增加一倍，可以避免70％的冗余计算，从而加速近35％。
Widely used Lagrangian numerical codes that compute the physical interaction with neighbouring resolution elements (particles), duplicate the calculation of the interaction between pairs of particles. We developed an algorithm that reduces the number of redundant calculations. The algorithm makes use of a hash function to flag already computed interactions and eventual collisions. The result of the hashing is stored in two caches. Without limiting the cache memory usage, all duplicated calculations can be avoided, achieving the speed-up of a factor on two. We show that, limiting the cache size (in bits) to double the typical number of neighbouring particles, 70 per cent of the redundant calculations can be avoided, yielding a speed-up of almost 35 per cent.
