|
| | GpuBndryFuncFab ()=default |
| | GpuBndryFuncFab (F const &a_f) |
| | GpuBndryFuncFab (F &&a_f) |
| void | operator() (Box const &bx, FArrayBox &dest, int dcomp, int numcomp, Geometry const &geom, Real time, const Vector< BCRec > &bcr, int bcomp, int orig_comp) |
| template<class FF> |
| void | ccfcdoit (Box const &bx, FArrayBox &dest, int dcomp, int numcomp, Geometry const &geom, Real time, const Vector< BCRec > &bcr, int bcomp, int orig_comp, FF const &fillfunc) |
| void | nddoit (Box const &bx, FArrayBox &dest, int dcomp, int numcomp, Geometry const &geom, Real time, const Vector< BCRec > &bcr, int bcomp, int orig_comp) |
template<class F>
class amrex::GpuBndryFuncFab< F >
In this gpu version, F is provided by the user. It needs to have a device operator() that can work on a cell/node for boundaries not handled by amrex::fab_filcc/fab_filnd. For face and nodal data, it is also called on the faces and nodes on the domain boundary.
For nodal data, and face data in 1D, ghosts get the boundary value before F is called: reflect/hoextrap BCs are not applied and reflect_odd does not zero the boundary value.