variables_mpi Module Reference

Variables associated with domain partitioning context. More...

Variables
integer	rank = 0
	Rank for the current local domain (associated to an MPI process).
type(mpi_comm)	mpi_comm_notus
	Notus MPI communicator (Cartesian aware). This variable is initialized to mpi_comm_world in the early stage of Notus initialization?
double precision, dimension(:), allocatable	global_domain_corners2_x
	Coordinates of the right, top, front corner.
double precision, dimension(:), allocatable	global_domain_corners2_y
double precision, dimension(:), allocatable, target	global_domain_corners2_z
Numbers of local domains.
Total number of local domains.
integer	n_mpi_proc = 1
integer	n_mpi_proc_x = 1
	Number of local domains along the x-axis.
integer	n_mpi_proc_y = 1
	Number of local domains along the y-axis.
integer	n_mpi_proc_z = 1
	Number of local domains along the z-axis.
integer, dimension(:), allocatable	proc_direction
	Values n_mpi_proc_x, n_mpi_proc_y, n_mpi_proc_z stored in an array.
integer, dimension(:), allocatable	proc_coordinate
	Coordinates of the current local domain in the Cartesian processor grid.
logical	is_repartitioning = .false.
	Will the domain be repartitioned?
logical	is_repartitioning_achieved = .false.
	Has the repartitioning process been already achieved?
logical	is_initial_step_repartitioning = .true.
	Is it the initial step in the repartitioning process? If so, stop the calculation and print the required number of processors.
integer	n_cells_proc = 0
	Number of cells per partition.
integer	label_wanted_partitioning = -1
	Label of the wanted partitioning.
Neighborhood information.
Domain ranks for the six neighbors. The value -1 means “no neighbor.” Give the rank of any domain from its relative coordinates.
integer, dimension(:,:,:), allocatable	neighbor_proc
integer, dimension(:), allocatable	neighbor_proc_list
	Give the list of the neighboring processors in the repartitioning process.
integer, dimension(:,:), allocatable	neighbor_points_list
	Give the start and end indices of the points to exchange between neighbor processors.
logical	mpi_exchange_full = .true.
	MPI exchange including corners (and edges in 3D)
integer	n_exchange_proc = -1
	Number of neighbor domains to communicate with.
Processor information with respect to boundaries.
False or true depending whether they are at a physical boundary or not Is the processor on the left boundary of the domain
logical	is_proc_on_left_boundary = .false.
logical	is_proc_on_right_boundary = .false.
	Is the processor on the right boundary of the domain.
logical	is_proc_on_bottom_boundary = .false.
	Is the processor on the bottom boundary of the domain.
logical	is_proc_on_top_boundary = .false.
	Is the processor on the top boundary of the domain.
logical	is_proc_on_back_boundary = .false.
	Is the processor on the back boundary of the domain.
logical	is_proc_on_front_boundary = .false.
	Is the processor on the front boundary of the domain.
Periodic information (location of the periodic boundary)
These variables are set ONLY IF there are more than one proc in the given direction. Example 1: 'is_proc_on_left_periodic == .false.' if 'is_periodic_x == .true.' and 'n_mpi_proc_x == 1'. Example 2: 'is_proc_on_left_periodic == .true.' if 'is_periodic_x == .true.' and 'n_mpi_proc_x > 1' and 'is_proc_on_left_boundary'.
logical	is_proc_on_left_periodic = .false.
logical	is_proc_on_right_periodic = .false.
logical	is_proc_on_top_periodic = .false.
logical	is_proc_on_bottom_periodic = .false.
logical	is_proc_on_back_periodic = .false.
logical	is_proc_on_front_periodic = .false.
Local domain start and end indices in the physical domain (with no ghost cells)
indexing.
integer	is_global_physical_domain = 1
integer	is_global_physical_domain_r = 1
integer	is_global_physical_domain_t = 1
integer	js_global_physical_domain = 1
integer	js_global_physical_domain_r = 1
integer	js_global_physical_domain_t = 1
integer	ks_global_physical_domain = 1
integer	ks_global_physical_domain_r = 1
integer	ks_global_physical_domain_t = 1
integer	ie_global_physical_domain = 1
integer	ie_global_physical_domain_r = 1
integer	ie_global_physical_domain_t = 1
integer	je_global_physical_domain = 1
integer	je_global_physical_domain_r = 1
integer	je_global_physical_domain_t = 1
integer	ke_global_physical_domain = 1
integer	ke_global_physical_domain_r = 1
integer	ke_global_physical_domain_t = 1
integer	isu_global_physical_domain = 1
integer	ieu_global_physical_domain = 1
integer	jsv_global_physical_domain = 1
integer	jev_global_physical_domain = 1
integer	ksw_global_physical_domain = 1
integer	kew_global_physical_domain = 1
Local domain start and end indices in the overlapping numerical domain (with boundary and overlapping boundary cells)
indexing.
integer	is_global_overlap_numerical_domain = 1
integer	is_global_overlap_numerical_domain_r = 1
integer	is_global_overlap_numerical_domain_t = 1
integer	js_global_overlap_numerical_domain = 1
integer	js_global_overlap_numerical_domain_r = 1
integer	js_global_overlap_numerical_domain_t = 1
integer	ks_global_overlap_numerical_domain = 1
integer	ks_global_overlap_numerical_domain_r = 1
integer	ks_global_overlap_numerical_domain_t = 1
integer	ie_global_overlap_numerical_domain = 1
integer	ie_global_overlap_numerical_domain_r = 1
integer	ie_global_overlap_numerical_domain_t = 1
integer	je_global_overlap_numerical_domain = 1
integer	je_global_overlap_numerical_domain_r = 1
integer	je_global_overlap_numerical_domain_t = 1
integer	ke_global_overlap_numerical_domain = 1
integer	ke_global_overlap_numerical_domain_r = 1
integer	ke_global_overlap_numerical_domain_t = 1
Coordinates of the left, bottom, back corner.
double precision, dimension(:), allocatable	global_domain_corners1_x
double precision, dimension(:), allocatable	global_domain_corners1_y
double precision, dimension(:), allocatable, target	global_domain_corners1_z
List of all partition sizes.
The arrays starts at zero en ends on n_mpi_proc_*-1. E.g.: global_domain_list_n*(0:n_mpi_proc_*-1)
integer, dimension(:), allocatable	global_domain_list_nx
integer, dimension(:), allocatable	global_domain_list_nx_r
integer, dimension(:), allocatable	global_domain_list_nx_t
integer, dimension(:), allocatable	global_domain_list_ny
integer, dimension(:), allocatable	global_domain_list_ny_r
integer, dimension(:), allocatable	global_domain_list_ny_t
integer, dimension(:), allocatable	global_domain_list_nz
integer, dimension(:), allocatable	global_domain_list_nz_r
integer, dimension(:), allocatable	global_domain_list_nz_t
List of the starting indices of the faces of all partitions.
The arrays starts at zero en ends on n_mpi_proc_*-1. E.g.: global_domain_list_*s*(0:n_mpi_proc_*-1)
integer, dimension(:), allocatable	global_domain_list_isu
integer, dimension(:), allocatable	global_domain_list_isu_r
integer, dimension(:), allocatable	global_domain_list_isu_t
integer, dimension(:), allocatable	global_domain_list_jsv
integer, dimension(:), allocatable	global_domain_list_jsv_r
integer, dimension(:), allocatable	global_domain_list_jsv_t
integer, dimension(:), allocatable	global_domain_list_ksw
integer, dimension(:), allocatable	global_domain_list_ksw_r
integer, dimension(:), allocatable	global_domain_list_ksw_t
List of the ending indices of the faces of all partitions.
The arrays starts at zero en ends on n_mpi_proc_*-1. E.g.: global_domain_list_*e*(0:n_mpi_proc_*-1)
integer, dimension(:), allocatable	global_domain_list_ieu
integer, dimension(:), allocatable	global_domain_list_ieu_r
integer, dimension(:), allocatable	global_domain_list_ieu_t
integer, dimension(:), allocatable	global_domain_list_jev
integer, dimension(:), allocatable	global_domain_list_jev_r
integer, dimension(:), allocatable	global_domain_list_jev_t
integer, dimension(:), allocatable	global_domain_list_kew
integer, dimension(:), allocatable	global_domain_list_kew_r
integer, dimension(:), allocatable	global_domain_list_kew_t

Detailed Description

The global domain is partitioned in a Cartesian arrangement of local domains, as described in computational_domain. The numbers of local domain per direction are n_mpi_proc_x, n_mpi_proc_y, and n_mpi_proc_z, and the total number local domains is n_mpi_proc = prox_nb_x × n_mpi_proc_y × n_mpi_proc_z. Each local domain is identified by its coordinates proc_coordinate in the partitioning system, and also by its rank. Both coordinates and ranks ranges from 0 to some n-1.

Figure 1: Local domain indexing.

The module variables are defined within the context of a local domain (SIMD paradigm); they are distributed or local variables. Although, some variables have the same value for each local domain, thus can be considered as a shared or global variable.

Global numerical domain.

As described in computational_domain, the mesh is indexed locally over an extended domain, the numerical domain. It is sometimes necessary to define a global indexing.

Refined grid (under development): grid can be globally refined for phase advection. Suffix _r is associated to the refined grids, suffix _t is associated to the temporary grid.