GPUによる相互相関PIVの高速化に関する研究

This study aims to develop a fast Particle Image Velocimetry (PIV) software by Graphics Processing Unit(GPU). This summary consists of 3 parts. In the first part, the methods to accelerate recursive direct/FFT-based cross-correlation PIV by GPU are discussed. After the comparison, GPU accelerated cross-correlation PIVs are implemented on GPU and about 800 times faster processing is achieved compared to using single CPU. In the second part, the methods to accelerate these PIVs by multiple GPUs (multi-GPU) are also discussed. By parallelizing single GPU accelerated PIV by multi-GPU, about 85% of the theoretical performance can be obtained compared to using single GPU. In the third part, recursive hybrid cross-correlation PIV scheme is proposed, which uses FFT-based method in the former step and use direct method in the latter step. Real PIV images are used to validate hybrid cross-correlation PIV and the advantages in terms of its range, precision, and time cost are shown