GPU-Parallel simulation of rigid fibers in Stokes flow
Uppsala University, Disciplinary Domain of Science and Technology, Mathematics and Computer Science, Department of Information Technology
Uppsala University, 2014
The simulation of a fiber suspension requires that all interactions between the fibers involved are computed. This is a compute-intensive N-body problem that is highly data parallel. Using the GPU for these types of computations can be very beneficial. In this thesis an extension to a simulator, written in MATLAB, for rigid fibers in Stokes flow was designed and implemented to improve the performance of the simulator. A part of the simulator responsible for computing these fiber-to-fiber interactions was ported to the GPU using the CUDA programming language dedicated to general-purpose computing on GPUs. To accomplish this an interface to MATLAB was created and the portion of code to be ported was parallelized and adapted in a way suitable to the GPU. The ported code proved to be 16 times faster than the original implementation.
June 25, 2014 by hgpu
