cp_fm_diag.F File Reference

#include "../base/base_uses.f90"

Go to the source code of this file.

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

Functions/Subroutines
subroutine, public	cp_fm_diag::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.
subroutine, public	cp_fm_diag::diag_finalize ()
	Finalize the diagonalization library.
subroutine, public	cp_fm_diag::choose_eigv_solver (matrix, eigenvectors, eigenvalues, info)
	Choose the Eigensolver depending on which library is available ELPA seems to be unstable for small systems.
subroutine, public	cp_fm_diag::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.
subroutine, public	cp_fm_diag::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
subroutine, public	cp_fm_diag::cp_fm_svd (matrix_a, matrix_eigvl, matrix_eigvr_t, eigval, info)
	decomposes a quadratic matrix into its singular value decomposition
subroutine, public	cp_fm_diag::cp_fm_power (matrix, work, exponent, threshold, n_dependent, verbose, eigvals)
	...
subroutine, public	cp_fm_diag::cp_fm_block_jacobi (matrix, eigenvectors, eigval, thresh, start_sec_block)
	...
subroutine, public	cp_fm_diag::cp_fm_geeig (amatrix, bmatrix, eigenvectors, eigenvalues, work)
	General Eigenvalue Problem AX = BXE Single option version: Cholesky decomposition of B.
subroutine, public	cp_fm_diag::cp_fm_geeig_canon (amatrix, bmatrix, eigenvectors, eigenvalues, work, epseig)
	General Eigenvalue Problem AX = BXE Use canonical diagonalization : U*s**(-1/2)

Variables
real(kind=dp), parameter, public	cp_fm_diag::eps_check_diag_default = 5.0E-14_dp
integer, save, public	cp_fm_diag::diag_type = 0
integer, parameter, public	cp_fm_diag::fm_diag_type_scalapack = 101
integer, parameter, public	cp_fm_diag::fm_diag_type_elpa = 102
integer, parameter, public	cp_fm_diag::fm_diag_type_cusolver = 103
integer, parameter, public	cp_fm_diag::fm_diag_type_dlaf = 104
integer, parameter, public	cp_fm_diag::fm_diag_type_default = FM_DIAG_TYPE_SCALAPACK