電磁界解析の手法の一つとしてFDTD (Finite Difference Time-Domain) 法は広く用いられているが，陽解法であるためCFL (Courant-Friedrichs-Lewy) 条件により時間離散間隔が制限を受ける.この制限を緩和し計算を高速化するために，CFL条件に対して無条件安定なLOD-FDTD (Locally One-Dimensional-FDTD) 法が提案された.一方，ハードウェアを用いた高速化の手法としてGPU (Graphics Processing Unit) を汎用計算に利用するGPGPU (General-purpose computing on GPU) が挙げられる.GPUはCPU (Central Processing Unit) と比ベプロセッサコア数が多いため，高速な並列計算が行えるメリットがある.本研究では，予備的検討として1次元LOD-FDTD法の陰的スキームの部分にJacobi法を用い，CPUコードとGPUコードの計算速度の比較を行った.その結果として，GPUを用いた計算ではCPUの場合に比べて約16倍高速化された.%Finite-difference time-domain (FDTD) method has been commonly used in the electromagnetic field analysis problems. However the time step is limited by the Courant-Friedrichs-Lewy (CFL) condition because of the nature of explicit scheme. To enhance the calculation speed by relaxing this restriction, locally one-dimensional (LOD) FDTD method has been proposed as an implicit scheme which is unconditionally stable to any time-step. Recently, the concept of general purpose computing on a graphics processing unit (GPGPU) was introduced to achieve high performance computing as one of the methods of hardware acceleration. The computation with GPU has advantage in parallel computing over the CPU computation, because it has extremely many core processors. In this study, the combination of the implicit method of LOD-FDTD and GPU acceleration is proposed as a preliminary investigation. In the 1D-L0D-FDTD code, classical Jacobi method is employed as a solver for simultaneous linear equations which appear in the implicit process part. The performance between CPU code and GPU code are compared. Consequently, calculation speed with GPU is 16 times faster than that with CPU.
展开▼
