Fix post merge and the changing of ParEnv to a pointer to the ParMatr…

…ix. We need to create (at least?) one parmatrix during interpolation. It needs to be created on matrix present on bulk in all parts. Not sure this is the best solution...

elmerice/Solvers/CalvingGeometry.F90

@@ -3491,14 +3491,15 @@ SUBROUTINE SwitchMesh(Model, Solver, OldMesh, NewMesh)
     TYPE(Matrix_t), POINTER :: WorkMatrix=>NULL()
     LOGICAL :: Found, Global, GlobalBubbles, Debug, DoPrevValues, &
          NoMatrix, DoOptimizeBandwidth, PrimaryVar, HasValuesInPartition, &
-         PrimarySolver
+         PrimarySolver,CreatedParMatrix
     LOGICAL, POINTER :: UnfoundNodes(:)=>NULL(), BulkUnfoundNodes(:)=>NULL()
-    INTEGER :: i,j,k,DOFs, nrows,n, dummyint
+    INTEGER :: i,j,k,DOFs, nrows,n, dummyint, ierr
     INTEGER, POINTER :: WorkPerm(:)=>NULL(), SolversToIgnore(:)=>NULL(), &
          SurfaceMaskPerm(:)=>NULL(), BottomMaskPerm(:)=>NULL()
     REAL(KIND=dp), POINTER :: WorkReal(:)=>NULL(), WorkReal2(:)=>NULL(), PArray(:,:) => NULL()
     REAL(KIND=dp) :: FrontOrientation(3), RotationMatrix(3,3), UnRotationMatrix(3,3), &
          globaleps, localeps
+    LOGICAL, ALLOCATABLE :: PartActive(:)
     CHARACTER(LEN=MAX_NAME_LEN) :: SolverName, WorkName
 
     INTERFACE
@@ -3557,6 +3558,10 @@ END SUBROUTINE InterpolateMeshToMesh
     !----------------------------------------------
 
     Var => OldMesh % Variables
+
+    ALLOCATE(PartActive(ParEnv % PEs))
+    CreatedParMatrix = .FALSE.
+
     DO WHILE( ASSOCIATED(Var) )
 
        DoPrevValues = ASSOCIATED(Var % PrevValues)
@@ -3635,6 +3640,9 @@ END SUBROUTINE InterpolateMeshToMesh
                 END IF
              END IF
 
+             IF(.NOT. CreatedParMatrix) &
+              CALL MPI_AllGather(.NOT. NoMatrix, 1, MPI_LOGICAL, PartActive, 1, MPI_LOGICAL, ELMER_COMM_WORLD, ierr)
+
              IF ( ASSOCIATED(Var % EigenValues) ) THEN
                 n = SIZE(Var % EigenValues)
 
@@ -3678,6 +3686,20 @@ END SUBROUTINE InterpolateMeshToMesh
              IF(ASSOCIATED(WorkSolver % Matrix)) CALL FreeMatrix(WorkSolver % Matrix)
              WorkSolver % Matrix => WorkMatrix
 
+             ! bit of a hack
+             ! since ParEnv become a pointer to ParMatrix we need to ensure one ParMatrix is formed
+             ! it needs to be from a solver present on all parts hence the all gather further up.
+             ! it seems we only need to this once per timestep/interpolation as ParEnv will have some thing
+             ! to point to. If we don't do this ParEnv % PEs, % MyPE etc. all become nans mucking eveything up!
+             IF ( ASSOCIATED(WorkSolver % Matrix) .and. ALL(PartActive) .and. .NOT. CreatedParMatrix) THEN
+                IF (.NOT. ASSOCIATED(WorkSolver % Matrix % ParMatrix) ) THEN
+                  WorkSolver % Mesh => NewMesh
+
+                  CALL ParallelInitMatrix( WorkSolver, WorkSolver % Matrix, WorkPerm)
+                  CreatedParMatrix = .TRUE.
+                END IF
+             END IF
+
              NULLIFY(WorkMatrix)
 
              !NOTE: We don't switch Solver % Variable here, because