Redistribution of DBCSR blocks and distributed matrix slices. More...

Functions/Subroutines
subroutine, public	cp_dbcsr_to_subgroups (mat_global, mat_local, para_env, num_pe_sub, block_row_ranges)
	...
subroutine, public	cp_dbcsr_from_subgroups (mat_local, mat_global, para_env, row_mapping, copy_only)
	Sum subgroup blocks into the global layout, or copy mapped blocks.
subroutine, public	cp_array_to_dbcsr_slices (array, nrows, ncols, para_env_sub, first_row, last_row, template, matrices, matrix_ptrs)
	Consume a distributed flattened array and create a DBCSR matrix per slice.

Detailed Description

Redistribution of DBCSR blocks and distributed matrix slices.

Function/Subroutine Documentation

subroutine, public cp_dbcsr_redistribution::cp_dbcsr_to_subgroups	(	type(dbcsr_type)	mat_global,
		type(dbcsr_type)	mat_local,
		type(mp_para_env_type), intent(in)	para_env,
		integer	num_pe_sub,
		integer, dimension(:, :), optional	block_row_ranges )

...

Parameters

Definition at line 53 of file cp_dbcsr_redistribution.F.

Here is the call graph for this function:

Here is the caller graph for this function:

subroutine, public cp_dbcsr_redistribution::cp_dbcsr_from_subgroups	(	type(dbcsr_type)	mat_local,
		type(dbcsr_type)	mat_global,
		type(mp_para_env_type), intent(in)	para_env,
		integer, dimension(:), intent(in), optional	row_mapping,
		logical, intent(in), optional	copy_only )

Sum subgroup blocks into the global layout, or copy mapped blocks.

Parameters

mat_local	...
mat_global	...
para_env	communicator covering all source and destination ranks
row_mapping	optional source-to-destination block-row permutation
copy_only	copy without summation, filtering, or consuming the source (default: false)

Definition at line 213 of file cp_dbcsr_redistribution.F.

Here is the call graph for this function:

Here is the caller graph for this function:

subroutine, public cp_dbcsr_redistribution::cp_array_to_dbcsr_slices	(	real(kind=dp), dimension(:, :), intent(inout), allocatable	array,
		integer, intent(in)	nrows,
		integer, intent(in)	ncols,
		type(mp_para_env_type), intent(in), pointer	para_env_sub,
		integer, intent(in)	first_row,
		integer, intent(in)	last_row,
		type(dbcsr_type), intent(inout)	template,
		type(dbcsr_type), dimension(:), intent(inout), optional, target	matrices,
		type(dbcsr_p_type), dimension(:), intent(inout), optional	matrix_ptrs )

Consume a distributed flattened array and create a DBCSR matrix per slice.

Parameters

array	local flattened rows; all ranks have the same number of slices; deallocated on return
nrows	number of rows in each output matrix
ncols	number of columns in each output matrix
para_env_sub	communicator for the output matrices
first_row	first flattened row held locally
last_row	last flattened row held locally
template	DBCSR template defining output distribution and block sizes
matrices	preallocated array of output matrices
matrix_ptrs	preallocated array of output pointers; pointees are allocated here

Note: Supply exactly one output representation. Flattened rows use (i-1)*ncols+j.

Definition at line 376 of file cp_dbcsr_redistribution.F.

Here is the call graph for this function:

Here is the caller graph for this function: