Suitable Formulations of Lagrange Remap Finite Volume Schemes for Manycore/GPU Architectures
This paper is dedicated to Lagrange-Remap schemes (also referred to as Lagrange-Euler schemes) and their suitable formulations for manycore/GPU architectures. High performance computing efficiency requires a suitable balance between floating point operations and memory accesses, uniform compactly supported stencils, memory alignment, SIMD-based instructions and minimal dereferencing into memory. We provide various formulations, from the basis geometrical remapping to remap by flux balances and operator splitting variant approach. We present numerical experiments of two-dimensional Euler hydrodynamics on Cartesian grids up to \(2048^2\) cells and provide performance results.
KeywordsRiemann Solver Pressure Wave Propagation Implicit Discretization Finite Volume Formalism Gradient Reconstruction
- 2.Herdman, J.A., Gaudin, W.P., et al.: Accelerating hydrocodes with OpenACC, OpenCL, and CUDA. In: 3rd International Workshop on Performance Modeling, Benchmarking and Simulation (PMBS12), held as part of the International Conference for High Performance Computing, Networking, Storage and Analysis (SC12) (2012)Google Scholar