hfx_communication Module Reference

Routines for data exchange between MPI processes. More...

Functions/Subroutines
subroutine, public	get_full_density (para_env, full_density, rho, number_of_p_entries, block_offset, kind_of, basis_parameter, get_max_vals_spin, rho_beta, antisymmetric)
	Collects full density matrix from all CPUs
subroutine, public	distribute_ks_matrix (para_env, full_ks, ks_matrix, number_of_p_entries, block_offset, kind_of, basis_parameter, off_diag_fac, diag_fac)
	Distributes the local full Kohn-Sham matrix to all CPUS
subroutine, public	scale_and_add_fock_to_ks_matrix (para_env, qs_env, ks_matrix, irep, scaling_factor)
	Distributes the local full Kohn-Sham matrix to all CPUS. Is called in case of adiabatic rescaling. This is just a refactored version of distribute_ks_matrix
subroutine, public	get_atomic_block_maps (matrix, basis_parameter, kind_of, is_assoc_atomic_block, number_of_p_entries, para_env, atomic_block_offset, set_offset, block_offset, map_atoms_to_cpus, nkind)
	create a several maps array that reflects the ks matrix sparsity

Detailed Description

Routines for data exchange between MPI processes.

History

04.2008 created [Manuel Guidon]

Author

Manuel Guidon

Function/Subroutine Documentation

get_full_density()

subroutine, public hfx_communication::get_full_density	(	type(mp_para_env_type), pointer	para_env,
		real(dp), dimension(:)	full_density,
		type(dbcsr_type), pointer	rho,
		integer, intent(in)	number_of_p_entries,
		integer, dimension(:), pointer	block_offset,
		integer, dimension(*)	kind_of,
		type(hfx_basis_type), dimension(:), pointer	basis_parameter,
		logical, intent(in)	get_max_vals_spin,
		type(dbcsr_type), optional, pointer	rho_beta,
		logical, intent(in)	antisymmetric )

• Collects full density matrix from all CPUs

Parameters

para_env	...
full_density	The full Density matrix
rho	Distributed density
number_of_p_entries	Maximal buffer size
block_offset	...
kind_of	...
basis_parameter	...
get_max_vals_spin	...
rho_beta	...
antisymmetric	...

History

11.2007 created [Manuel Guidon]

Author

Manuel Guidon

Note

• Communication with left/right node only added a mpi_sync before and after the ring of isendrecv. This speed up the communication, and might protect against idle neighbors flooding a busy node with messages [Joost]

Definition at line 73 of file hfx_communication.F.

Here is the call graph for this function:

Here is the caller graph for this function:

distribute_ks_matrix()

subroutine, public hfx_communication::distribute_ks_matrix	(	type(mp_para_env_type), pointer	para_env,
		real(dp), dimension(:)	full_ks,
		type(dbcsr_type), pointer	ks_matrix,
		integer, intent(in)	number_of_p_entries,
		integer, dimension(:), pointer	block_offset,
		integer, dimension(*)	kind_of,
		type(hfx_basis_type), dimension(:), pointer	basis_parameter,
		real(dp), intent(in), optional	off_diag_fac,
		real(dp), intent(in), optional	diag_fac )

• Distributes the local full Kohn-Sham matrix to all CPUS

Parameters

para_env	...
full_ks	The full Kohn-Sham matrix
ks_matrix	Distributed Kohn-Sham matrix
number_of_p_entries	Maximal buffer size
block_offset	...
kind_of	...
basis_parameter	...
off_diag_fac	...
diag_fac	...

History

11.2007 created [Manuel Guidon]

Author

Manuel Guidon

Note

• Communication with left/right node only

Definition at line 200 of file hfx_communication.F.

Here is the call graph for this function:

Here is the caller graph for this function:

scale_and_add_fock_to_ks_matrix()

subroutine, public hfx_communication::scale_and_add_fock_to_ks_matrix	(	type(mp_para_env_type), pointer	para_env,
		type(qs_environment_type), pointer	qs_env,
		type(dbcsr_p_type), dimension(:, :), pointer	ks_matrix,
		integer, intent(in)	irep,
		real(dp), intent(in)	scaling_factor )

• Distributes the local full Kohn-Sham matrix to all CPUS. Is called in case of adiabatic rescaling. This is just a refactored version of distribute_ks_matrix

Parameters

para_env	...
qs_env	...
ks_matrix	Distributed Kohn-Sham matrix
irep	...
scaling_factor	...

History

11.2007 created [Manuel Guidon]

Author

Manuel Guidon

Note

• Communication with left/right node only

Definition at line 307 of file hfx_communication.F.

Here is the call graph for this function:

Here is the caller graph for this function:

get_atomic_block_maps()

subroutine, public hfx_communication::get_atomic_block_maps	(	type(dbcsr_type), pointer	matrix,
		type(hfx_basis_type), dimension(:)	basis_parameter,
		integer, dimension(:)	kind_of,
		integer, dimension(:, :), intent(out)	is_assoc_atomic_block,
		integer, intent(out)	number_of_p_entries,
		type(mp_para_env_type), pointer	para_env,
		integer, dimension(:, :), pointer	atomic_block_offset,
		integer, dimension(:, :, :, :), pointer	set_offset,
		integer, dimension(:), pointer	block_offset,
		type(hfx_2d_map), dimension(:), pointer	map_atoms_to_cpus,
		integer	nkind )

create a several maps array that reflects the ks matrix sparsity

Parameters

matrix	...
basis_parameter	...
kind_of	...
is_assoc_atomic_block	...
number_of_p_entries	...
para_env	...
atomic_block_offset	...
set_offset	...
block_offset	...
map_atoms_to_cpus	...
nkind	...

History

11.2007 refactored [Joost VandeVondele] 07.2009 add new maps

Author

Manuel Guidon \notes is_assoc_atomic_block returns the mpi rank + 1 for associated blocks, zero for unassiated blocks

Definition at line 423 of file hfx_communication.F.

Here is the call graph for this function:

Here is the caller graph for this function: