m_domain_data_exchange Module Reference

Module for data exchange between neighbouring subdomains in a domain decomposition. For distribution of TensorProdFun accross subdomains, see the distribute() TBP in m_tensorprod_basis.f90. More...

Functions/Subroutines
subroutine, public	data_exchange (domain, neighbours, data_blocks, move_to)
	Subroutine to exchange blocks of data between neighbouring subdomains.
subroutine, public	guard_layer_offset (offset, tp_space, x, y, z, node_based)
	Determine the neighbour offset for a particle at position (x, y, z).

Variables
integer, parameter, public	remove_block = -huge(1)

Detailed Description

Module for data exchange between neighbouring subdomains in a domain decomposition. For distribution of TensorProdFun accross subdomains, see the distribute() TBP in m_tensorprod_basis.f90.

Function/Subroutine Documentation

data_exchange()

subroutine, public m_domain_data_exchange::data_exchange	(	class(domaindecomp), intent(in)	domain,
		type(tensorprodneighbour), dimension(-max_comm_neighbours_x:max_comm_neighbours_x, -max_comm_neighbours_y:max_comm_neighbours_y, -max_comm_neighbours_z:max_comm_neighbours_z), intent(in)	neighbours,
		real(wp), dimension(:, :), intent(inout), allocatable	data_blocks,
		integer, dimension(:, :), intent(in)	move_to )

Subroutine to exchange blocks of data between neighbouring subdomains.

Parameters

[in]	domain	The local DomainDecomp of the current rank
[in]	neighbours	The array of neighbouring TensorProdNeighbour objects
[in,out]	data_blocks	The blocks of data to be sent/received. A 2D array where the second index is the block index
[in]	move_to	An array indicating the direction to move each block of data. A 2D array where the first index is the direction (1=x, 2=y, 3=z) and the second index is the block index

Note

For example, suppose we want to move one block of reals to the positive x-direction neighbour, and one block to the negative y-direction neighbour. Then we would set move_to(:, 1) = [1, 0, 0] and move_to(:, 2) = [0, -1, 0], and data_blocks(:, 1) and data_blocks(:, 2) would contain the data to be sent in each case. [0, 0, 0] keeps the data on the current rank. At the end of the subroutine, data_blocks(:, :) contains the received data from the neighbours, as well as the data that was kept on the current rank.

guard_layer_offset()

subroutine, public m_domain_data_exchange::guard_layer_offset	(	integer, dimension(3), intent(out)	offset,
		type(tensorprodspace), intent(in)	tp_space,
		real(wp), intent(in)	x,
		real(wp), intent(in)	y,
		real(wp), intent(in)	z,
		logical, intent(in), optional	node_based )

Determine the neighbour offset for a particle at position (x, y, z).

Given a position in physical space, this subroutine determines which neighbour (if any) the particle should be sent to, expressed as a 3D offset (ri, rj, rk) in the neighbour array. If the particle is within our own responsible intervals, offset = [0, 0, 0]. The search is split into three independent 1D searches.

Parameters

[out]	offset	The 3D neighbour offset (ri, rj, rk)
[in]	tp_space	The TensorProdSpace
[in]	x	The x-coordinate of the particle
[in]	y	The y-coordinate of the particle
[in]	z	The z-coordinate of the particle
[in]	node_based	Whether the guard layer is node-based (true) or rank-based (false) (default: .false.)