DBCSR operations in CP2K. More...

Data Types
type	dbcsr_fm_copy_plan_type
interface	dbcsr_allocate_matrix_set
interface	dbcsr_deallocate_matrix_set

Functions/Subroutines
subroutine, public	copy_fm_to_dbcsr (fm, matrix, keep_sparsity)
	Copy a BLACS matrix to a dbcsr matrix.
subroutine, public	copy_fm_to_dbcsr_bc (fm, bc_mat)
	Copy a BLACS matrix to a dbcsr matrix with a special block-cyclic distribution, which requires no complete redistribution.
subroutine, public	copy_dbcsr_to_fm (matrix, fm, plan, check_plan)
	Copy a DBCSR matrix to a BLACS matrix.
subroutine, public	copy_dbcsr_to_fm_bc (bc_mat, fm)
	Copy a DBCSR_BLACS matrix to a BLACS matrix.
subroutine, public	dbcsr_copy_columns_hack (matrix_b, matrix_a, ncol, source_start, target_start, para_env, blacs_env)
	hack for dbcsr_copy_columns
subroutine, public	cp_dbcsr_dist2d_to_dist (dist2d, dist)
	Creates a DBCSR distribution from a distribution_2d.
subroutine, public	dbcsr_multiply_local (matrix_a, vec_b, vec_c, ncol, alpha)
	multiply a dbcsr with a replicated array c = alpha_scalar * A (dbscr) * b + c
subroutine, public	cp_dbcsr_sm_fm_multiply (matrix, fm_in, fm_out, ncol, alpha, beta)
	multiply a dbcsr with a fm matrix
subroutine, public	cp_dbcsr_plus_fm_fm_t (sparse_matrix, matrix_v, matrix_g, ncol, alpha, keep_sparsity, symmetry_mode)
	performs the multiplication sparse_matrix+dense_mat*dens_mat^T if matrix_g is not explicitly given, matrix_v^T will be used this can be important to save the necessary redistribute for a different matrix_g and increase performance.
subroutine, public	cp_fm_to_dbcsr_row_template (matrix, fm_in, template)
	Utility function to copy a specially shaped fm to dbcsr_matrix The result matrix will be the matrix in dbcsr format with the row blocks sizes according to the block_sizes of the template and the col blocks sizes evenly blocked with the internal dbcsr conversion size (32 is the current default).
subroutine, public	cp_dbcsr_m_by_n_from_template (matrix, template, m, n, sym)
	Utility function to create an arbitrary shaped dbcsr matrix with the same processor grid as the template matrix both row sizes and col sizes are evenly blocked with the internal dbcsr_conversion size (32 is the current default).
subroutine, public	cp_dbcsr_m_by_n_from_row_template (matrix, template, n, sym)
	Utility function to create dbcsr matrix, m x n matrix (n arbitrary) with the same processor grid and row distribution as the template matrix col sizes are evenly blocked with the internal dbcsr_conversion size (32 is the current default).
subroutine, public	dbcsr_create_dist_r_unrot (dist_right, dist_left, ncolumns, right_col_blk_sizes)
	Creates a new distribution for the right matrix in a matrix multiplication with unrotated grid.

Variables
integer, save, public	max_elements_per_block = 32

Detailed Description

DBCSR operations in CP2K.

Author: Urban Borstnik

Date: 2009-05-12

Version: 0.8

Modification history:

Created 2009-05-12
Generalized sm_fm_mulitply for matrices w/ different row/col block size (A. Bussy, 11.2018)

Function/Subroutine Documentation

◆ copy_fm_to_dbcsr()

subroutine, public cp_dbcsr_operations::copy_fm_to_dbcsr	(	type(cp_fm_type), intent(in)	fm,
		type(dbcsr_type), intent(inout)	matrix,
		logical, intent(in), optional	keep_sparsity )

Copy a BLACS matrix to a dbcsr matrix.

    real_matrix=beta*real_matrix+alpha*fm
    beta defaults to 0, alpha to 1

Parameters

[in]	fm	full matrix
[out]	matrix	DBCSR matrix
[in]	keep_sparsity	(optional) retains the sparsity of the input matrix

Date: 2009-10-13

History: 2009-10-13 rewritten based on copy_dbcsr_to_fm

Author: Urban Borstnik

Version: 2.0

Definition at line 128 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

◆ copy_fm_to_dbcsr_bc()

subroutine, public cp_dbcsr_operations::copy_fm_to_dbcsr_bc	(	type(cp_fm_type), intent(in)	fm,
		type(dbcsr_type), intent(inout)	bc_mat )

Copy a BLACS matrix to a dbcsr matrix with a special block-cyclic distribution, which requires no complete redistribution.

Parameters

fm	...
bc_mat	...

Definition at line 166 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ copy_dbcsr_to_fm()

subroutine, public cp_dbcsr_operations::copy_dbcsr_to_fm	(	type(dbcsr_type), intent(in)	matrix,
		type(cp_fm_type), intent(inout)	fm,
		type(dbcsr_fm_copy_plan_type), intent(inout), optional	plan,
		logical, intent(in), optional	check_plan )

Copy a DBCSR matrix to a BLACS matrix.

Parameters

[in]	matrix	DBCSR matrix
[out]	fm	full matrix
	plan	optional S/A routing plan; discard before releasing the FM structure
	check_plan	check for layout changes and collectively rebuild the plan; defaults to false for callers that guarantee fixed layouts and threads

Definition at line 232 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

◆ copy_dbcsr_to_fm_bc()

subroutine, public cp_dbcsr_operations::copy_dbcsr_to_fm_bc	(	type(dbcsr_type), intent(in)	bc_mat,
		type(cp_fm_type), intent(inout)	fm )

Copy a DBCSR_BLACS matrix to a BLACS matrix.

Parameters

	bc_mat	DBCSR matrix
[out]	fm	full matrix

Definition at line 553 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ dbcsr_copy_columns_hack()

subroutine, public cp_dbcsr_operations::dbcsr_copy_columns_hack	(	type(dbcsr_type), intent(inout)	matrix_b,
		type(dbcsr_type), intent(in)	matrix_a,
		integer, intent(in)	ncol,
		integer, intent(in)	source_start,
		integer, intent(in)	target_start,
		type(mp_para_env_type), pointer	para_env,
		type(cp_blacs_env_type), pointer	blacs_env )

hack for dbcsr_copy_columns

Parameters

matrix_b	...
matrix_a	...
ncol	...
source_start	...
target_start	...
para_env	...
blacs_env	...

Author: vw

Definition at line 650 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_dbcsr_dist2d_to_dist()

subroutine, public cp_dbcsr_operations::cp_dbcsr_dist2d_to_dist	(	type(distribution_2d_type), intent(in), target	dist2d,
		type(dbcsr_distribution_type), intent(out)	dist )

Creates a DBCSR distribution from a distribution_2d.

Parameters

[in]	dist2d	distribution_2d
[out]	dist	DBCSR distribution

History: move form dbcsr_operation 01.2010

Definition at line 695 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ dbcsr_multiply_local()

subroutine, public cp_dbcsr_operations::dbcsr_multiply_local	(	type(dbcsr_type), intent(in)	matrix_a,
		real(dp), dimension(:, :), intent(in)	vec_b,
		real(dp), dimension(:, :), intent(inout)	vec_c,
		integer, intent(in), optional	ncol,
		real(dp), intent(in), optional	alpha )

multiply a dbcsr with a replicated array c = alpha_scalar * A (dbscr) * b + c

Parameters

[in]	matrix_a	DBSCR matrxx
[in]	vec_b	vectors b
[in,out]	vec_c	vectors c
[in]	ncol	nbr of columns
[in]	alpha	alpha

Definition at line 735 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

◆ cp_dbcsr_sm_fm_multiply()

subroutine, public cp_dbcsr_operations::cp_dbcsr_sm_fm_multiply	(	type(dbcsr_type), intent(in)	matrix,
		type(cp_fm_type), intent(in)	fm_in,
		type(cp_fm_type), intent(inout)	fm_out,
		integer, intent(in)	ncol,
		real(dp), intent(in), optional	alpha,
		real(dp), intent(in), optional	beta )

multiply a dbcsr with a fm matrix

For backwards compatibility with BLAS XGEMM, this routine supports the multiplication of matrices with incompatible dimensions.

Parameters

[in]	matrix	DBCSR matrix
	fm_in	full matrix
	fm_out	full matrix
[in]	ncol	nbr of columns
[in]	alpha	alpha
[in]	beta	beta

Definition at line 828 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

◆ cp_dbcsr_plus_fm_fm_t()

subroutine, public cp_dbcsr_operations::cp_dbcsr_plus_fm_fm_t	(	type(dbcsr_type), intent(inout)	sparse_matrix,
		type(cp_fm_type), intent(in)	matrix_v,
		type(cp_fm_type), intent(in), optional	matrix_g,
		integer, intent(in)	ncol,
		real(kind=dp), intent(in), optional	alpha,
		logical, intent(in), optional	keep_sparsity,
		integer, intent(in), optional	symmetry_mode )

performs the multiplication sparse_matrix+dense_mat*dens_mat^T if matrix_g is not explicitly given, matrix_v^T will be used this can be important to save the necessary redistribute for a different matrix_g and increase performance.

Parameters

sparse_matrix	...
matrix_v	...
matrix_g	...
ncol	...
alpha	...
keep_sparsity	Determines if the sparsity of sparse_matrix is retained by default it is TRUE
symmetry_mode	There are the following modes 1: sparse_matrix += 0.5alpha(vg^T+g^Tv) (symmetric update) -1: sparse_matrix += 0.5alpha(vg^T-g^Tv) (skewsymmetric update) else: sparse_matrix += alphavg^T (no symmetry, default) saves some redistribution steps

Definition at line 973 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_fm_to_dbcsr_row_template()

subroutine, public cp_dbcsr_operations::cp_fm_to_dbcsr_row_template	(	type(dbcsr_type), intent(inout)	matrix,
		type(cp_fm_type), intent(in)	fm_in,
		type(dbcsr_type), intent(in)	template )

Utility function to copy a specially shaped fm to dbcsr_matrix The result matrix will be the matrix in dbcsr format with the row blocks sizes according to the block_sizes of the template and the col blocks sizes evenly blocked with the internal dbcsr conversion size (32 is the current default).

Parameters

matrix	...
fm_in	...
template	...

Definition at line 1138 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_dbcsr_m_by_n_from_template()

subroutine, public cp_dbcsr_operations::cp_dbcsr_m_by_n_from_template	(	type(dbcsr_type), intent(inout)	matrix,
		type(dbcsr_type), intent(inout)	template,
		integer, intent(in)	m,
		integer, intent(in)	n,
		character, intent(in), optional	sym )

Utility function to create an arbitrary shaped dbcsr matrix with the same processor grid as the template matrix both row sizes and col sizes are evenly blocked with the internal dbcsr_conversion size (32 is the current default).

Parameters

matrix	dbcsr matrix to be created
template	template dbcsr matrix giving its mp_env
m	global row size of output matrix
n	global col size of output matrix
sym	...

Definition at line 1172 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_dbcsr_m_by_n_from_row_template()

subroutine, public cp_dbcsr_operations::cp_dbcsr_m_by_n_from_row_template	(	type(dbcsr_type), intent(inout)	matrix,
		type(dbcsr_type), intent(inout)	template,
		integer	n,
		character, optional	sym )

Utility function to create dbcsr matrix, m x n matrix (n arbitrary) with the same processor grid and row distribution as the template matrix col sizes are evenly blocked with the internal dbcsr_conversion size (32 is the current default).

Parameters

matrix	dbcsr matrix to be created
template	template dbcsr matrix giving its mp_env
n	global col size of output matrix
sym	...

Definition at line 1215 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ dbcsr_create_dist_r_unrot()

subroutine, public cp_dbcsr_operations::dbcsr_create_dist_r_unrot	(	type(dbcsr_distribution_type), intent(out)	dist_right,
		type(dbcsr_distribution_type), intent(in)	dist_left,
		integer, intent(in)	ncolumns,
		integer, dimension(:), intent(out), pointer	right_col_blk_sizes )

Creates a new distribution for the right matrix in a matrix multiplication with unrotated grid.

Parameters

[out]	dist_right	new distribution for the right matrix
[in]	dist_left	the distribution of the left matrix
[in]	ncolumns	number of columns in right matrix
[out]	right_col_blk_sizes	sizes of blocks in the created column

The new row distribution for the right matrix is the same as the row: distribution of the left matrix, while the column distribution is created so that it is appropriate to the parallel environment.

Definition at line 1363 of file cp_dbcsr_operations.F.

Here is the call graph for this function:

Here is the caller graph for this function:

Variable Documentation

◆ max_elements_per_block

integer, save, public cp_dbcsr_operations::max_elements_per_block = 32

Definition at line 58 of file cp_dbcsr_operations.F.

Data Types

Functions/Subroutines

Variables

Detailed Description

Function/Subroutine Documentation

◆ copy_fm_to_dbcsr()

◆ copy_fm_to_dbcsr_bc()

◆ copy_dbcsr_to_fm()

◆ copy_dbcsr_to_fm_bc()

◆ dbcsr_copy_columns_hack()

◆ cp_dbcsr_dist2d_to_dist()

◆ dbcsr_multiply_local()

◆ cp_dbcsr_sm_fm_multiply()

◆ cp_dbcsr_plus_fm_fm_t()

◆ cp_fm_to_dbcsr_row_template()

◆ cp_dbcsr_m_by_n_from_template()

◆ cp_dbcsr_m_by_n_from_row_template()

◆ dbcsr_create_dist_r_unrot()

Variable Documentation

◆ max_elements_per_block