Functions
void	blacs_exit_ (const int *continue_mpi)

void	Cblacs_gridexit (int context)

void	Cblacs_gridinfo (int context, int nprow, int npcol, int myrow, int mycol)

void	Cblacs_gridinit (int context, const char layout, int nprow, int npcol)

void	Cfree_blacs_system_handle (int system_handle)

int	Csys2blacs_handle (MPI_Comm mpi_comm)

void	descinit_ (int desc, const int m, const int n, const int mb, const int nb, const int irsrc, const int icsrc, const int ictxt, const int lld, int info)

int	numroc_ (const int n, const int nb, const int iproc, const int isrcproc, const int *nprocs)

void	pzhegvx_ (const int ibtype, const char jobz, const char range, const char uplo, const int n, std::complex< double > a, const int ia, const int ja, const int desca, std::complex< double > b, const int ib, const int jb, const int descb, const double vl, const double vu, const int il, const int iu, const double abstol, int m, int nz, double w, const double orfac, std::complex< double > z, const int iz, const int jz, const int descz, std::complex< double > work, const int lwork, double rwork, const int lrwork, int iwork, const int liwork, int ifail, int iclustr, double gap, int info)

Detailed Description

ScaLAPACK declarations for distributed complex double-precision routines.

Provides the Fortran interfaces for BLACS grid management and distributed eigensolvers (PZHEGVX).

Function Documentation

◆ blacs_exit_()

void q3::scalapack::blacs_exit_ ( const int * continue_mpi )

BLACS: Exit from BLACS.

◆ Cblacs_gridexit()

void q3::scalapack::Cblacs_gridexit ( int context )

◆ Cblacs_gridinfo()

void q3::scalapack::Cblacs_gridinfo	(	int	context,
		int *	nprow,
		int *	npcol,
		int *	myrow,
		int *	mycol
	)

◆ Cblacs_gridinit()

void q3::scalapack::Cblacs_gridinit	(	int *	context,
		const char *	layout,
		int	nprow,
		int	npcol
	)

◆ Cfree_blacs_system_handle()

void q3::scalapack::Cfree_blacs_system_handle ( int system_handle )

◆ Csys2blacs_handle()

int q3::scalapack::Csys2blacs_handle ( MPI_Comm mpi_comm )

BLACS grid management, C entry points.

These must not be mixed with the Fortran blacs_grid*_ entry points. Csys2blacs_handle returns a handle in the C layer's numbering space, and handing that to Fortran blacs_gridinit_ makes BLACS read it as a communicator: it returned 1 here, and MPICH then aborted inside MPI_Comm_group with "Invalid communicator, comm=0x1". Going through Cblacs_gridinit instead gives each communicator its own grid correctly.

Csys2blacs_handle takes a C MPI_Comm, hence <mpi.h> above; typing it int would compile only where MPI_Comm is an integer (MPICH yes, OpenMPI no). Note also that Cblacs_gridinfo and Cblacs_gridexit take the context BY VALUE, unlike their Fortran counterparts which take a pointer.

The system handle from Csys2blacs_handle is a separate resource from the grid context Cblacs_gridinit derives from it, and each needs releasing: Cblacs_gridexit for the grid, Cfree_blacs_system_handle for the handle.

◆ descinit_()

void q3::scalapack::descinit_	(	int *	desc,
		const int *	m,
		const int *	n,
		const int *	mb,
		const int *	nb,
		const int *	irsrc,
		const int *	icsrc,
		const int *	ictxt,
		const int *	lld,
		int *	info
	)

DESCINIT: Initialize a ScaLAPACK descriptor.

◆ numroc_()

int q3::scalapack::numroc_	(	const int *	n,
		const int *	nb,
		const int *	iproc,
		const int *	isrcproc,
		const int *	nprocs
	)

NUMROC: Compute number of rows/cols on a local process.

◆ pzhegvx_()

void q3::scalapack::pzhegvx_	(	const int *	ibtype,
		const char *	jobz,
		const char *	range,
		const char *	uplo,
		const int *	n,
		std::complex< double > *	a,
		const int *	ia,
		const int *	ja,
		const int *	desca,
		std::complex< double > *	b,
		const int *	ib,
		const int *	jb,
		const int *	descb,
		const double *	vl,
		const double *	vu,
		const int *	il,
		const int *	iu,
		const double *	abstol,
		int *	m,
		int *	nz,
		double *	w,
		const double *	orfac,
		std::complex< double > *	z,
		const int *	iz,
		const int *	jz,
		const int *	descz,
		std::complex< double > *	work,
		const int *	lwork,
		double *	rwork,
		const int *	lrwork,
		int *	iwork,
		const int *	liwork,
		int *	ifail,
		int *	iclustr,
		double *	gap,
		int *	info
	)

PZHEGVX: Parallel generalized Hermitian-definite eigenvalue problem.

The expert driver, because it is the only one ScaLAPACK offers for the generalized problem. Q3 previously declared and called pzhegv_, the simple driver by analogy with LAPACK's ZHEGV – but ScaLAPACK has no PZHEGV. The API provides PZHEEV, PZHEEVD and PZHEEVX for the standard problem and PZHEGVX alone for the generalized one, so every ScaLAPACK-enabled build failed to link with ‘undefined reference to pzhegv_’ against any conforming implementation, Cray libsci included.

Beyond PZHEGV's arguments this takes the eigenvalue selection (range, vl, vu, il, iu, abstol), returns how many it found (m) and how many eigenvectors it computed (nz), takes a reorthogonalization threshold (orfac), and needs integer workspace plus the ifail, iclustr and gap diagnostic arrays.

Functions