RGX_Create_Task_GridS

Creates Regular Sub-Grid, one for each process. The first routine to be called after MPI_Init(),

Synopsis


int RGX_Create_Task_GridS(
                           MPI_Comm  World_Comm,
                           int       Ndim,

                           int      *N_task_ptr,
                           int       N_tasks_1dim[],
                           MPI_Comm *Cart_Grids_Comm_ptr
                         )

Unmodified Input Variables

World_Comm - Input Communicator containing all processors for the grid (Usually MPI_COMM_WORLD)
Ndim - Dimensionality of the Grid (Integer)

Modified Output Variables

N_task_ptr - Pointer to Number of Tasks in the Communicator
N_tasks_1dim - Integer Array, of size Ndim, of Number of Tasks Per Dimension in the Grid. It is implicitly assumed to be N_tasks_1dim[ Ndim ].
Cart_Grids_Comm_ptr - Pointer to the Cartesian Grid Communicator
returned value - returns MPI_SUCCESS if completed successfully

Notes on Fortran Interface

The RGX routines which have this message have a Fortran interface. If the C function that is defined here has a returned integer status, the Fortran interface will have an additional argument, ierr, at the end of its argument list.

Notes on Minimal Set of Variables

In order to use RGX for a regualr grid calculation. There is a minimal set of variables needed to be used with the RGX_XXX routines. The set of variables is listed as follows,


   const int     MD_type = MPI_DOUBLE_PRECISION;
   double       *Task_xGrid;         // Data of Local Task Grid              
   int           Ndim;               // Dimensionality of the 
   //                                // grid           
   MPI_Comm      RGX_Comm;           // Communicator of all 
   //                                // sub-grids        
   int           N_task;             // No. of process/tasks 
   //                                // in Communicator 
   int          *Glb_Grid_sz;        // Global grid sizes :
   //                                // Glb_Grid_sz(Ndim) 
   int          *task_Ngbrs;         // the tid of the nearest
   //                                // neighbor task :
   //                                // task_Ngbrs(2, Ndim)                
   int          *task_Grid_endpts;   // local sub-grid's end
   //                                // points :
   //                                // task_Grid_endpts(Ndim, 2)          
   int          *N_task_1dims;       // No of task 
   //                                // per dimension :             
   //                                // N_task_1dims(Ndim)                 
   int          *stride_width;       // width of the stride of
   //                                // ghost point :
   //                                // stride_width(Ndim)                 
   MPI_Datatype *task_bndry_strides; // local task/grid's 
   //                                // boundary strides :
   //                                // task_bndry_strides(Ndim)           

Memory for each of these pointer variables can be allocated dynammically/statically, but it has to be continuous.

Supported Data Type of the Block Array

There are currently 5 datatypes supported in RGX. The types are identified by MPI_Datatype. They are MPI_INTEGER, MPI_REAL, MPI_DOUBLE_PRECISION, MPI_COMPLEX and MPI_DOUBLE_COMPLEX.

Definition Location

This subroutine is defined in the librgx.a.

Location:../src/librgx/RGX_Create_Task_GridS.c