used for collecting some of the diagonalization schemes available for cp_fm_type. cp_fm_power also moved here as it is very related More...

Functions/Subroutines
subroutine, public	diag_init (diag_lib, fallback_applied, elpa_kernel, elpa_neigvec_min_input, elpa_qr, elpa_print, elpa_qr_unsafe, dlaf_neigvec_min_input, eps_check_diag_input)
	Setup the diagonalization library to be used. More...

subroutine, public	diag_finalize ()
	Finalize the diagonalization library. More...

subroutine, public	choose_eigv_solver (matrix, eigenvectors, eigenvalues, info)
	Choose the Eigensolver depending on which library is available ELPA seems to be unstable for small systems. More...

subroutine, public	cp_fm_syevd (matrix, eigenvectors, eigenvalues, info)
	Computes all eigenvalues and vectors of a real symmetric matrix significantly faster than syevx, scales also much better. Needs workspace to allocate all the eigenvectors. More...

subroutine, public	cp_fm_syevx (matrix, eigenvectors, eigenvalues, neig, work_syevx)
	compute eigenvalues and optionally eigenvectors of a real symmetric matrix using scalapack. If eigenvectors are required this routine will replicate a full matrix on each CPU... if more than a handful of vectors are needed, use cp_fm_syevd instead More...

subroutine, public	cp_fm_power (matrix, work, exponent, threshold, n_dependent, verbose, eigvals)
	... More...

subroutine, public	cp_fm_block_jacobi (matrix, eigenvectors, eigval, thresh, start_sec_block)
	... More...

subroutine, public	cp_fm_geeig (amatrix, bmatrix, eigenvectors, eigenvalues, work)
	General Eigenvalue Problem AX = BXE Single option version: Cholesky decomposition of B. More...

subroutine, public	cp_fm_geeig_canon (amatrix, bmatrix, eigenvectors, eigenvalues, work, epseig)
	General Eigenvalue Problem AX = BXE Use canonical diagonalization : Us*(-1/2) More...

Variables
real(kind=dp), parameter, public	eps_check_diag_default = 5.0E-14_dp

integer, parameter, public	fm_diag_type_scalapack = 101

integer, parameter, public	fm_diag_type_elpa = 102

integer, parameter, public	fm_diag_type_cusolver = 103

integer, parameter, public	fm_diag_type_dlaf = 104

integer, parameter, public	fm_diag_type_default = FM_DIAG_TYPE_SCALAPACK

Detailed Description

used for collecting some of the diagonalization schemes available for cp_fm_type. cp_fm_power also moved here as it is very related

Note: first version : most routines imported

History

unused Jacobi routines removed, cosmetics (05.04.06,MK)

Author: Joost VandeVondele (2003-08)

Function/Subroutine Documentation

◆ diag_init()

subroutine, public cp_fm_diag::diag_init	(	character(len=*), intent(in)	diag_lib,
		logical, intent(out)	fallback_applied,
		integer, intent(in)	elpa_kernel,
		integer, intent(in)	elpa_neigvec_min_input,
		logical, intent(in)	elpa_qr,
		logical, intent(in)	elpa_print,
		logical, intent(in)	elpa_qr_unsafe,
		integer, intent(in)	dlaf_neigvec_min_input,
		real(kind=dp), intent(in)	eps_check_diag_input
	)

Setup the diagonalization library to be used.

Parameters

diag_lib	diag_library flag from GLOBAL section in input
fallback_applied	.TRUE. if support for the requested library was not compiled-in and fallback to ScaLAPACK was applied, .FALSE. otherwise.
elpa_kernel	integer that determines which ELPA kernel to use for diagonalization
elpa_neigvec_min_input	...
elpa_qr	logical that determines if ELPA should try to use QR to accelerate the diagonalization procedure of suitably sized matrices
elpa_print	logical that determines if information about the ELPA diagonalization should be printed
elpa_qr_unsafe	logical that enables potentially unsafe ELPA options
dlaf_neigvec_min_input	...
eps_check_diag_input	...

History

Add support for DLA-Future (05.09.2023, RMeli)

Author: MI 11.2013

Definition at line 132 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ diag_finalize()

subroutine, public cp_fm_diag::diag_finalize

Finalize the diagonalization library.

Definition at line 187 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ choose_eigv_solver()

subroutine, public cp_fm_diag::choose_eigv_solver	(	type(cp_fm_type), intent(in)	matrix,
		type(cp_fm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		integer, intent(out), optional	info
	)

Choose the Eigensolver depending on which library is available ELPA seems to be unstable for small systems.

Parameters

matrix	...
eigenvectors	...
eigenvalues	...
info	...

info If present returns error code and prevents program stops.: Works currently only for cp_fm_syevd with ScaLAPACK. Other solvers will end the program regardless of PRESENT(info).

History

Do not use ELPA for small matrices and use instead ScaLAPACK as fallback (10.05.2021, MK)

Definition at line 207 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_fm_syevd()

subroutine, public cp_fm_diag::cp_fm_syevd	(	type(cp_fm_type), intent(in)	matrix,
		type(cp_fm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		integer, intent(out), optional	info
	)

Computes all eigenvalues and vectors of a real symmetric matrix significantly faster than syevx, scales also much better. Needs workspace to allocate all the eigenvectors.

Parameters

matrix	...
eigenvectors	...
eigenvalues	...
info	...

matrix is supposed to be in upper triangular form, and overwritten by this routine

info If present returns error code and prevents program stops.: Works currently only for scalapack. Other solvers will end the program regardless of PRESENT(info).

Definition at line 412 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_fm_syevx()

subroutine, public cp_fm_diag::cp_fm_syevx	(	type(cp_fm_type), intent(in)	matrix,
		type(cp_fm_type), intent(in), optional	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		integer, intent(in), optional	neig,
		real(kind=dp), intent(in), optional	work_syevx
	)

compute eigenvalues and optionally eigenvectors of a real symmetric matrix using scalapack. If eigenvectors are required this routine will replicate a full matrix on each CPU... if more than a handful of vectors are needed, use cp_fm_syevd instead

Parameters

matrix	...
eigenvectors	...
eigenvalues	...
neig	...
work_syevx	...

matrix is supposed to be in upper triangular form, and overwritten by this routine: neig is the number of vectors needed (default all) work_syevx evec calculation only, is the fraction of the working buffer allowed (1.0 use full buffer) reducing this saves time, but might cause the routine to fail

Definition at line 656 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_fm_power()

subroutine, public cp_fm_diag::cp_fm_power	(	type(cp_fm_type), intent(in)	matrix,
		type(cp_fm_type), intent(in)	work,
		real(kind=dp), intent(in)	exponent,
		real(kind=dp), intent(in)	threshold,
		integer, intent(out)	n_dependent,
		logical, intent(in), optional	verbose,
		real(kind=dp), dimension(2), intent(out), optional	eigvals
	)

...

Parameters

matrix	...
work	...
exponent	...
threshold	...
n_dependent	...
verbose	...
eigvals	...

Definition at line 895 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_fm_block_jacobi()

subroutine, public cp_fm_diag::cp_fm_block_jacobi	(	type(cp_fm_type), intent(in)	matrix,
		type(cp_fm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(in)	eigval,
		real(kind=dp), intent(in)	thresh,
		integer, intent(in)	start_sec_block
	)

...

Parameters

matrix	...
eigenvectors	...
eigval	...
thresh	...
start_sec_block	...

Definition at line 1073 of file cp_fm_diag.F.

Here is the caller graph for this function:

◆ cp_fm_geeig()

subroutine, public cp_fm_diag::cp_fm_geeig	(	type(cp_fm_type), intent(in)	amatrix,
		type(cp_fm_type), intent(in)	bmatrix,
		type(cp_fm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:)	eigenvalues,
		type(cp_fm_type), intent(in)	work
	)

General Eigenvalue Problem AX = BXE Single option version: Cholesky decomposition of B.

Parameters

amatrix	...
bmatrix	...
eigenvectors	...
eigenvalues	...
work	...

Definition at line 1273 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

◆ cp_fm_geeig_canon()

subroutine, public cp_fm_diag::cp_fm_geeig_canon	(	type(cp_fm_type), intent(in)	amatrix,
		type(cp_fm_type), intent(in)	bmatrix,
		type(cp_fm_type), intent(in)	eigenvectors,
		real(kind=dp), dimension(:), intent(out)	eigenvalues,
		type(cp_fm_type), intent(in)	work,
		real(kind=dp), intent(in)	epseig
	)

General Eigenvalue Problem AX = BXE Use canonical diagonalization : U*s**(-1/2)

Parameters

amatrix	...
bmatrix	...
eigenvectors	...
eigenvalues	...
work	...
epseig	...

Definition at line 1315 of file cp_fm_diag.F.

Here is the call graph for this function:

Here is the caller graph for this function:

Variable Documentation

◆ eps_check_diag_default

real(kind=dp), parameter, public cp_fm_diag::eps_check_diag_default = 5.0E-14_dp

Definition at line 69 of file cp_fm_diag.F.

◆ fm_diag_type_scalapack

integer, parameter, public cp_fm_diag::fm_diag_type_scalapack = 101

Definition at line 87 of file cp_fm_diag.F.

◆ fm_diag_type_elpa

integer, parameter, public cp_fm_diag::fm_diag_type_elpa = 102

Definition at line 87 of file cp_fm_diag.F.

◆ fm_diag_type_cusolver

integer, parameter, public cp_fm_diag::fm_diag_type_cusolver = 103

Definition at line 87 of file cp_fm_diag.F.

◆ fm_diag_type_dlaf

integer, parameter, public cp_fm_diag::fm_diag_type_dlaf = 104

Definition at line 87 of file cp_fm_diag.F.

◆ fm_diag_type_default

integer, parameter, public cp_fm_diag::fm_diag_type_default = FM_DIAG_TYPE_SCALAPACK

Definition at line 98 of file cp_fm_diag.F.

Functions/Subroutines

Variables

Detailed Description

Function/Subroutine Documentation

◆ diag_init()

◆ diag_finalize()

◆ choose_eigv_solver()

◆ cp_fm_syevd()

◆ cp_fm_syevx()

◆ cp_fm_power()

◆ cp_fm_block_jacobi()

◆ cp_fm_geeig()

◆ cp_fm_geeig_canon()

Variable Documentation

◆ eps_check_diag_default

◆ fm_diag_type_scalapack

◆ fm_diag_type_elpa

◆ fm_diag_type_cusolver

◆ fm_diag_type_dlaf

◆ fm_diag_type_default