DPIVSoft-OpenCL: a multicore CPU-GPU accelerated open source code for 2D Particle Image Velocimetry

Abstract

We present a translation of the original Matlab DPIVSoft code to a complete open source code implemented in Python, to perform Particle Image Velocimetry (PIV) in two-dimensions, in parallel, and with interrogation window shifting along with the double-pass window deformation approach using multiple iterations for each pass. The added value of the code is the use of the Open Computing Language (OpenCL) library to parallelize the original code on multiple Intel Central Processing Units (CPUs) and/or Graphics Processing Units (GPUs), so it can be run on all commercially available GPUs. Examples of flow application are included in the text using synthetic images generated from DNS data from John Hopkins Turbulence Database (JHTD) (Perlman, 2007), showing about 90x speedup over the previous Matlab implementation for a given test case.