Implementation of HyPar on GPU is a work-in-progress. Currently, only the following parts of the code has been implemented on NVidia GPUs with CUDA:

Spatial discretization schemes:
- 5th order WENO scheme (Interp1PrimFifthOrderWENO())
Time integration schemes:
- Runge-Kutta time integration (TimeRK())
Boundary conditions:
- Periodic (_PERIODIC_)
- "Slip-wall" (_SLIP_WALL_)
Physical models:
- 3D Navier-Stokes (NavierStokes3D)

Implementation note:

The CUDA implementation uses a different variable ordering than the CPU code. While the CPU code uses the "Array of Structures" variable ordering, where the innermost loop is over the vector components at a grid point, the CUDA implementation uses the "Structure of Arrays" ordering, where the innermost loops are over the grid points.
This implementation is designed for 1 GPU per MPI rank. Thus, on a node, the simulation should be run with as many MPI ranks as there are GPUs.