Bug in analytic functions
Analytic functions as input MeshFunction in Hamilton-Jacobi solver with MPI on GPU give bad values.
Generated function from tnl-init is passed into tnl-direct-eikonal-solver. TNL devides input MeshFunction into blocks, which number depends on number of processes. Devided MeshFunction that we get in file tnlDirectEikonalProblem_impl.h in function Solve() has invalid values.
Starting script is attached.