task_list_methods Module Reference

generate the tasks lists used by collocate and integrate routines More...

Functions/Subroutines
subroutine, public	generate_qs_task_list (ks_env, task_list, reorder_rs_grid_ranks, skip_load_balance_distributed, soft_valid, basis_type, pw_env_external, sab_orb_external)
	...
subroutine, public	task_list_inner_loop (tasks, ntasks, curr_tasks, rs_descs, dft_control, cube_info, gridlevel_info, cindex, iatom, jatom, rpgfa, rpgfb, zeta, zetb, kind_radius_b, set_radius_a, set_radius_b, ra, rab, la_max, la_min, lb_max, lb_min, npgfa, npgfb, nseta, nsetb)
	...
subroutine, public	distribute_tasks (rs_descs, ntasks, natoms, tasks, atom_pair_send, atom_pair_recv, symmetric, reorder_rs_grid_ranks, skip_load_balance_distributed)
	Assembles tasks to be performed on local grid.
subroutine, public	rs_distribute_matrix (rs_descs, pmats, atom_pair_send, atom_pair_recv, nimages, scatter, hmats)
	redistributes the matrix so that it can be used in realspace operations i.e. according to the task lists for collocate and integrate. This routine can become a bottleneck in large calculations.
subroutine, public	rs_scatter_matrices (src_matrices, dest_buffer, task_list, group)
	Scatters dbcsr matrix blocks and receives them into a buffer as needed before collocation.
subroutine, public	rs_gather_matrices (src_buffer, dest_matrices, task_list, group)
	Gather the dbcsr matrix blocks and receives them into a buffer as needed after integration.
subroutine, public	rs_copy_to_buffer (src_matrices, dest_buffer, task_list)
	Copies the DBCSR blocks into buffer, replaces rs_scatter_matrix for non-distributed grids.
subroutine, public	rs_copy_to_matrices (src_buffer, dest_matrices, task_list)
	Copies from buffer into DBCSR matrics, replaces rs_gather_matrix for non-distributed grids.

Detailed Description

generate the tasks lists used by collocate and integrate routines

History

01.2008 [Joost VandeVondele] refactered out of qs_collocate / qs_integrate

Author

Joost VandeVondele

Function/Subroutine Documentation

generate_qs_task_list()

subroutine, public task_list_methods::generate_qs_task_list	(	type(qs_ks_env_type), pointer	ks_env,
		type(task_list_type), pointer	task_list,
		logical, intent(in)	reorder_rs_grid_ranks,
		logical, intent(in)	skip_load_balance_distributed,
		logical, intent(in), optional	soft_valid,
		character(len=*), intent(in), optional	basis_type,
		type(pw_env_type), optional, pointer	pw_env_external,
		type(neighbor_list_set_p_type), dimension(:), optional, pointer	sab_orb_external
	)

...

Parameters

ks_env	...
task_list	...
reorder_rs_grid_ranks	Flag that indicates if this routine should or should not overwrite the rs descriptor (see comment below)
skip_load_balance_distributed	...
soft_valid	...
basis_type	...
pw_env_external	...
sab_orb_external	...

History

01.2008 factored out of calculate_rho_elec [Joost VandeVondele] 04.2010 divides tasks into grid levels and atom pairs for integrate/collocate [Iain Bethune] (c) The Numerical Algorithms Group (NAG) Ltd, 2010 on behalf of the HECToR project 06.2015 adjusted to be used with multiple images (k-points) [JGH]

Note

If this routine is called several times with different task lists, the default behaviour is to re-optimize the grid ranks and overwrite the rs descriptor and grids. reorder_rs_grid_ranks = .FALSE. prevents the code of performing a new optimization by leaving the rank order in its current state.

Definition at line 116 of file task_list_methods.F.

Here is the call graph for this function:

Here is the caller graph for this function:

task_list_inner_loop()

subroutine, public task_list_methods::task_list_inner_loop	(	type(task_type), dimension(:), pointer	tasks,
		integer	ntasks,
		integer	curr_tasks,
		type(realspace_grid_desc_p_type), dimension(:), pointer	rs_descs,
		type(dft_control_type), pointer	dft_control,
		type(cube_info_type), dimension(:), pointer	cube_info,
		type(gridlevel_info_type), pointer	gridlevel_info,
		integer	cindex,
		integer	iatom,
		integer	jatom,
		real(kind=dp), dimension(:, :), pointer	rpgfa,
		real(kind=dp), dimension(:, :), pointer	rpgfb,
		real(kind=dp), dimension(:, :), pointer	zeta,
		real(kind=dp), dimension(:, :), pointer	zetb,
		real(kind=dp)	kind_radius_b,
		real(kind=dp), dimension(:), pointer	set_radius_a,
		real(kind=dp), dimension(:), pointer	set_radius_b,
		real(kind=dp), dimension(3)	ra,
		real(kind=dp), dimension(3)	rab,
		integer, dimension(:), pointer	la_max,
		integer, dimension(:), pointer	la_min,
		integer, dimension(:), pointer	lb_max,
		integer, dimension(:), pointer	lb_min,
		integer, dimension(:), pointer	npgfa,
		integer, dimension(:), pointer	npgfb,
		integer	nseta,
		integer	nsetb
	)

...

Parameters

tasks	...
ntasks	...
curr_tasks	...
rs_descs	...
dft_control	...
cube_info	...
gridlevel_info	...
cindex	...
iatom	...
jatom	...
rpgfa	...
rpgfb	...
zeta	...
zetb	...
kind_radius_b	...
set_radius_a	...
set_radius_b	...
ra	...
rab	...
la_max	...
la_min	...
lb_max	...
lb_min	...
npgfa	...
npgfb	...
nseta	...
nsetb	...

History

Joost VandeVondele: 10.2008 refactored

Definition at line 660 of file task_list_methods.F.

Here is the call graph for this function:

Here is the caller graph for this function:

distribute_tasks()

subroutine, public task_list_methods::distribute_tasks	(	type(realspace_grid_desc_p_type), dimension(:), pointer	rs_descs,
		integer	ntasks,
		integer	natoms,
		type(task_type), dimension(:), pointer	tasks,
		type(atom_pair_type), dimension(:), pointer	atom_pair_send,
		type(atom_pair_type), dimension(:), pointer	atom_pair_recv,
		logical, intent(in)	symmetric,
		logical, intent(in)	reorder_rs_grid_ranks,
		logical, intent(in)	skip_load_balance_distributed
	)

Assembles tasks to be performed on local grid.

Parameters

rs_descs	the grids
ntasks	Number of tasks for local processing
natoms	...
nimages	...
tasks	the task set generated on this processor
rval	...
atom_pair_send	...
atom_pair_recv	...
symmetric	...
reorder_rs_grid_ranks	...
skip_load_balance_distributed	...

History

none

Author

MattW 21/11/2007

Definition at line 1898 of file task_list_methods.F.

Here is the call graph for this function:

Here is the caller graph for this function:

rs_distribute_matrix()

subroutine, public task_list_methods::rs_distribute_matrix	(	type(realspace_grid_desc_p_type), dimension(:), pointer	rs_descs,
		type(dbcsr_p_type), dimension(:), pointer	pmats,
		type(atom_pair_type), dimension(:), pointer	atom_pair_send,
		type(atom_pair_type), dimension(:), pointer	atom_pair_recv,
		integer	nimages,
		logical	scatter,
		type(dbcsr_p_type), dimension(:), optional, pointer	hmats
	)

redistributes the matrix so that it can be used in realspace operations i.e. according to the task lists for collocate and integrate. This routine can become a bottleneck in large calculations.

Parameters

rs_descs	...
pmats	...
atom_pair_send	...
atom_pair_recv	...
natoms	...
nimages	...
scatter	...
hmats	...

Definition at line 2185 of file task_list_methods.F.

Here is the call graph for this function:

Here is the caller graph for this function:

rs_scatter_matrices()

subroutine, public task_list_methods::rs_scatter_matrices	(	type(dbcsr_p_type), dimension(:), intent(in)	src_matrices,
		type(offload_buffer_type), intent(inout)	dest_buffer,
		type(task_list_type), intent(in)	task_list,
		type(mp_comm_type), intent(in)	group
	)

Scatters dbcsr matrix blocks and receives them into a buffer as needed before collocation.

Author

Ole Schuett

Definition at line 2524 of file task_list_methods.F.

Here is the caller graph for this function:

rs_gather_matrices()

subroutine, public task_list_methods::rs_gather_matrices	(	type(offload_buffer_type), intent(in)	src_buffer,
		type(dbcsr_p_type), dimension(:), intent(inout)	dest_matrices,
		type(task_list_type), intent(in)	task_list,
		type(mp_comm_type), intent(in)	group
	)

Gather the dbcsr matrix blocks and receives them into a buffer as needed after integration.

Author

Ole Schuett

Definition at line 2559 of file task_list_methods.F.

Here is the caller graph for this function:

rs_copy_to_buffer()

subroutine, public task_list_methods::rs_copy_to_buffer	(	type(dbcsr_p_type), dimension(:), intent(in)	src_matrices,
		type(offload_buffer_type), intent(inout)	dest_buffer,
		type(task_list_type), intent(in)	task_list
	)

Copies the DBCSR blocks into buffer, replaces rs_scatter_matrix for non-distributed grids.

Author

Ole Schuett

Definition at line 2595 of file task_list_methods.F.

Here is the caller graph for this function:

rs_copy_to_matrices()

subroutine, public task_list_methods::rs_copy_to_matrices	(	type(offload_buffer_type), intent(in)	src_buffer,
		type(dbcsr_p_type), dimension(:), intent(inout)	dest_matrices,
		type(task_list_type), intent(in)	task_list
	)

Copies from buffer into DBCSR matrics, replaces rs_gather_matrix for non-distributed grids.

Author

Ole Schuett

Definition at line 2611 of file task_list_methods.F.

Here is the call graph for this function:

Here is the caller graph for this function: