循环分块技术可以提高Cache的命中率,而循环分块的规模是该技术的决定性因素。利用粒子群优化算法可以优化循环分块的规模,获取合适规模的分块,提高循环分块的效率。论文首先研究了LU分解算法,然后讨论了传统粒子群优化算法并针对用于循环分块方面的不足加以改进,最后把优化的粒子群优化算法用于LU分解算法之中,从而提出了一个PSO-LU循环分块算法。仿真实验结果表明,和原始基准测试程序相,所提循环分块算法比性能有所提升,其更充分地提高高速缓存运算效率。
Loop tiling can be used to improve hit rate of the cache. The scale of tiling is the key factor of the technology. The neural network model of Particle Swarm Optimization algorithm is applied to optimize the scale of tiling. The technology can obtain a more suitable tiling scale than the classical tiling algorithm, and result in improving the efficiency of tiling. Paper first studied the LU decomposition algorithm, and then discuses the traditional particle swarm optimization algorithm which was improved in light of the shortcomings of the for loop block. Finally, the improved particle swarm optimization algorithm was used to LU decomposition algorithm and a PSO - LU Loop tiling algorithm was proposed. Compared to the original benchmark achieve, we proposed using LU decomposition loop tiling algorithm to improve the operation efficiency of Cache.