GPUによる粒子法半陰解法アルゴリズムの高速化  [in Japanese] SEMI-IMPLICIT ALGORITHM OF PARTICLE METHOD ACCELERATED BY GPU  [in Japanese]

Access this Article

Author(s)

Abstract

粒子法は複雑な自由表面の変化の追跡に適していることから,水工学分野でも有用性が示され,海岸工学においては数値波動水槽への応用も進められている.しかし計算負荷が高いため,ハードウェアへの相応の投資が必要となり,これが一般利用の障害となっている.ところで,画像処理目的に開発されてきたGPUを,一般的な学術演算の高速化に利用する取り組みが最近活発となっている.本研究では,MPS法を高速化する手段のひとつとして,CUDAによるGPU併用コードを開発した.高速化を達成するための留意点を整理し,MPS法計算コード固有の特性に適合するように移植を行った.メモリ配置の工夫に加えて,計算のコアとなる近傍粒子探索や圧力のPoisson方程式の収束計算については,GPU計算を効率化するため,特に綿密に検討した.

  The MPS (Moving Particle Semi-implicit) method has been proven useful in free-surface hydrodynamic flows. Despite its applicability, one of its practical problems is the high computational load. By the way, Graphics Processing Unit (GPU), which was originally developed for computer graphics acceleration, now provides unprecedented computational capability for scientific applications. </br>  The main objective of this study is to develop a GPU-accelerated MPS code with using CUDA (Compute Unified Device Architecture) language. Several techniques have been shown to optimize calculations in CUDA. In order to promote the acceleration by GPU, particular attentions are given to both the search of neighboring particles and the iterative solution of simultaneous linear equations in the Poisson Pressure Equation. In this paper, dam break flow calculations have been carried out by the GPU-accelerated MPS method, and the accuracy and performance of GPU-based code are investigated by comparing the results with those by CPU.

Journal

  • Doboku Gakkai Ronbunshuu B

    Doboku Gakkai Ronbunshuu B 66(2), 217-222, 2010

    Japan Society of Civil Engineers

Codes

Page Top