lapack_interface.c File Reference

interface methods for eigenvector computation and matrix multiplication using openblas More...

Go to the source code of this file.

Macros
#define	BMS_CALL(x)
#define	SCIP_RealTOINT(x)   ((LAPACKINTTYPE) (x + 0.5))

Typedefs
typedef int	LAPACKINTTYPE

Functions
Functions
SCIP_RETCODE	SCIPlapackComputeEigenvectorDecomposition (BMS_BUFMEM *bufmem, int n, SCIP_Real *A, SCIP_Real *eigenvalues, SCIP_Real *eigenvectors)
SCIP_RETCODE	SCIPlapackMatrixVectorMult (int nrows, int ncols, SCIP_Real *matrix, SCIP_Real *vector, SCIP_Real *result)
SCIP_RETCODE	SCIPlapackMatrixMatrixMult (int nrowsA, int ncolsA, SCIP_Real *matrixA, SCIP_Bool transposeA, int nrowsB, int ncolsB, SCIP_Real *matrixB, SCIP_Bool transposeB, SCIP_Real *result)
SCIP_RETCODE	SCIPlapackLinearSolve (BMS_BUFMEM *bufmem, int m, int n, SCIP_Real *A, SCIP_Real *b, SCIP_Real *x)

Detailed Description

interface methods for eigenvector computation and matrix multiplication using openblas

Author

Sonja Mars

Lars Schewe

Tristan Gally

Marc Pfetsch

This file is used to call the LAPACK routine DSYEVR (double-symmetric-eigenvector computation) and the BLAS routine DGEMV (double-general-matrix-vector multiplication).

If a version of openblas/lapack with 64 bit integers is used in connection with SDPA, then one should define LAPACK_LONGLONGINT. Note that one cannot detect the precision with which openblas/lapack is compiled, so this define is essential for correct functioning (resulting in a segmentation fault otherwise).

Definition in file lapack_interface.c.

Macro Definition Documentation

#define BMS_CALL

(

x

)

Value:

do                                                                                      \
                      {                                                                                       \
                          if( NULL == (x) )                                                                   \
                          {                                                                                   \
                             SCIPerrorMessage("No memory in function call\n");                                \
                             return SCIP_NOMEMORY;                                                            \
                          }                                                                                   \
                      }                                                                                       \
                      while( FALSE )

Checks if a BMSallocMemory-call was successfull, otherwise returns SCIP_NOMEMORY

Definition at line 73 of file lapack_interface.c.

Referenced by SCIPlapackComputeEigenvectorDecomposition(), and SCIPlapackLinearSolve().

#define SCIP_RealTOINT

(

x

)

((LAPACKINTTYPE) (x + 0.5))

transforms a SCIP_Real (that should be integer, but might be off by some numerical error) to an integer by adding an epsilon and rounding down

Definition at line 84 of file lapack_interface.c.

Referenced by SCIPlapackComputeEigenvectorDecomposition(), and SCIPlapackLinearSolve().

Typedef Documentation

typedef int LAPACKINTTYPE

Definition at line 68 of file lapack_interface.c.

Function Documentation

SCIP_RETCODE SCIPlapackComputeEigenvectorDecomposition	(	BMS_BUFMEM *	bufmem,
		int	n,
		SCIP_Real *	A,
		SCIP_Real *	eigenvalues,
		SCIP_Real *	eigenvectors
	)

computes the eigenvector decomposition of a symmetric matrix using LAPACK

Parameters

bufmem	buffer memory
n	size of matrix
A	matrix for which the decomposition should be computed
eigenvalues	pointer to store eigenvalues (should be length n)
eigenvectors	pointer to store eigenvectors (should be length n*n), eigenvectors are given as rows

Definition at line 250 of file lapack_interface.c.

References BMS_CALL, F77_FUNC, and SCIP_RealTOINT.

Referenced by calcRelax(), and SCIP_DECL_CONSCHECK().

SCIP_RETCODE SCIPlapackMatrixVectorMult	(	int	nrows,
		int	ncols,
		SCIP_Real *	matrix,
		SCIP_Real *	vector,
		SCIP_Real *	result
	)

performs matrix-vector-multiplication using BLAS

Parameters

nrows	number of rows in matrix
ncols	number of cols in matrix
matrix	the matrix we want to multiply
vector	vector we want to multiply with the matrix
result	pointer to store the resulting vector

Definition at line 354 of file lapack_interface.c.

References F77_FUNC.

Referenced by cutUsingEigenvector().

SCIP_RETCODE SCIPlapackMatrixMatrixMult	(	int	nrowsA,
		int	ncolsA,
		SCIP_Real *	matrixA,
		SCIP_Bool	transposeA,
		int	nrowsB,
		int	ncolsB,
		SCIP_Real *	matrixB,
		SCIP_Bool	transposeB,
		SCIP_Real *	result
	)

performs matrix-matrix-multiplication A*B using BLAS

Parameters

nrowsA	number of rows in matrix A
ncolsA	number of cols in matrix A
matrixA	matrix A given as nrowsA * ncolsA array
transposeA	Should matrix A be transposed before multiplication?
nrowsB	number of rows in matrix B
ncolsB	number of cols in matrix B
matrixB	matrix B given as ncolsA * ncolsB array
transposeB	Should matrix B be transposed before multiplication?
result	pointer to nrowsA * nrowsB array to store the resulting matrix

Definition at line 393 of file lapack_interface.c.

References F77_FUNC.

Referenced by calcRelax(), and SCIP_DECL_CONSCHECK().

SCIP_RETCODE SCIPlapackLinearSolve	(	BMS_BUFMEM *	bufmem,
		int	m,
		int	n,
		SCIP_Real *	A,
		SCIP_Real *	b,
		SCIP_Real *	x
	)

computes the minimum-norm solution to a real linear least squares problem: minimize 2-norm(| b - A*x |) using LAPACK (uses singular value decomposition of A). A is an M-by-N matrix which may be rank-deficient.

Parameters

bufmem	buffer memory
m	number of rows of A
n	number of columns of A
A	coefficient matrix of the linear system
b	right-hand side of the linear system (should be length max(m,n))
x	pointer to store values for x (should be length n)

Definition at line 438 of file lapack_interface.c.

References BMS_CALL, F77_FUNC, and SCIP_RealTOINT.

Referenced by SCIP_DECL_CONSCHECK().