high performance computing on graphics processing units: hgpu.org

The progress of high performance computing platforms is dramatic, and most of the simulations carried out on these platforms, result in improvements on one level, yet exposes shortcomings of the current CFD packages capabilities. Therefore, hardware-aware design and optimizations are crucial towards exploiting the modern computing resources. This thesis proposes optimizations aimed at acceleration numerical simulations, which is illustrated in OpenFOAM solvers. A hybrid MPI and GPGPU parallel conjugate gradient linear solver has been designed and implemented to solve the sparse linear algebraic kernel that derived from two CFD solvers: icoFoam, which is an incompressible flow solver, and laplacianFoam, which solves the Laplace equation. A load-balancing step is applied by using heterogeneous decomposition, which decompose the computations in proportion to the performance of each computing device and take into account minimizing the communications. In addition, we implemented the pipeline conjugate gradient solver as an algorithmic improvement, and parallelized it using: MPI, GPGPU and hybrid. The experimental results show that the proposed hybrid implementation of both laplacianFoam and icoFoam significantly outperform the state-of-the-art implementations.