加速并行时域有限差分仿真的新方法

New acceleration technique for parallel FDTD simulation

为了加速并行时域有限差分仿真,提出了基于单指令多数据流式扩展(SSE)的一种新的加速方法,在Intel T2300的PC机上实现了对并行时域有限差分仿真的加速,给出了基于消息传递接口(MPI)、OpenMP和SSE指令集的三级数据并行算法。为了验证该算法的加速效率,计算了自由空间中电磁波的辐射问题,得到的加速比为2.62。实验结果表明:这种加速技术无需任何额外的硬件投资就能在很大程度上提高计算效率。

To accelerate the simulation of parallel finite-difference time-domain （FDTD）, this paper proposes a new numerical acceleration technique based on the streaming SIMD extensions（SSE）, and gives an implementation on Intel T2300 platform, and then puts forward a three-level data parallel algorithm based on mes- sage passing interface（MPI）, OpenMP and SSE. The numerical experiments for the electromagnetic radiation problems have demonstrated the high efficiency of this new technique, and can achieve an acceleration rate of 2.62. The result shows that this technique dramatically improves the computing efficiency without any extra hardware investment.