Developer Reference

Intel® oneAPI Math Kernel Library LAPACK Examples

ID 766877
Date 3/22/2024
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

C Interface Conventions

The C interfaces are implemented for most of the oneMKL LAPACK driver and computational routines.

The arguments of the C interfaces for the oneMKL LAPACK functions comply with the following rules:

  • Scalar input arguments are passed by value.

  • Array arguments are passed by reference.

  • Array input arguments are declared with the const modifier.

  • Function arguments are passed by pointer.

  • An integer return value replaces the info output parameter. The return value equal to 0 means the function operation is completed successfully. See the section Error Codes below.

  • Most of the LAPACK C interfaces have an additional parameter matrix_layout of the int type as their first argument. This parameter specifies whether the two-dimensional arrays are row-major (LAPACK_ROW_MAJOR) or column-major (LAPACK_COL_MAJOR).

The LAPACK C interface omits workspace parameters because workspace is allocated at run time and released upon completion of the function operation.

For the ?gesvd function, work arrays contain valuable information on exit. The interface contains an additional argument superb.

Function Types

The function types are used in non-symmetric eigenproblem functions only.

typedef lapack_logical (*LAPACK_S_SELECT2) (const float*, const float*);
typedef lapack_logical (*LAPACK_S_SELECT3) (const float*, const float*, const float*);
typedef lapack_logical (*LAPACK_D_SELECT2) (const double*, const double*);
typedef lapack_logical (*LAPACK_D_SELECT3) (const double*, const double*, const double*);
typedef lapack_logical (*LAPACK_C_SELECT1) (const lapack_complex_float*);
typedef lapack_logical (*LAPACK_C_SELECT2) (const lapack_complex_float*, const lapack_complex_float*);
typedef lapack_logical (*LAPACK_Z_SELECT1) (const lapack_complex_double*);
typedef lapack_logical (*LAPACK_Z_SELECT2) (const lapack_complex_double*, const lapack_complex_double*);

Mapping FORTRAN Data Types against C Data Types

FORTRAN Data Types C Data Types
INTEGER lapack_int
LOGICAL lapack_logical
REAL float
DOUBLE PRECISION double
COMPLEX lapack_complex_float
COMPLEX*16 lapack_complex_double
CHARACTER char

C Types

#ifndef lapack_int
#define lapack_int MKL_INT
#endif
#ifndef lapack_logical
#define lapack_logical lapack_int
#endif

Complex Types

For single precision:

#ifndef lapack_complex_float
#define lapack_complex_float   MKL_Complex8
#endif

For double precision:

#ifndef lapack_complex_double
#define lapack_complex_double   MKL_Complex16
#endif

Matrix Layout

Most of the LAPACK C interfaces have an additional parameter matrix_layout of type int as their first argument. This parameter specifies whether the two-dimensional arrays are row-major LAPACK_ROW_MAJOR or column-major LAPACK_COL_MAJOR.

In general the leading dimension lda is equal to the number of elements in the major dimension. It is also equal to the distance in elements between two neighboring elements in a line in the minor dimension. If there are no extra elements in a matrix with m rows and n columns, then

  • For row-major ordering: the number of elements in a row is n, and row i is stored in memory right after row i-1. Therefore lda is n.
  • For column-major ordering: the number of elements in a column is m, and column i is stored in memory right after column i-1. Therefore lda is m.

To refer to a submatrix with dimensions k by l, use the number of elements in the major dimension of the whole matrix (as above) as the leading dimension and k and l in the subroutine's input parameters to describe the size of the submatrix.


submatrix example

Matrix Ordering

#define LAPACK_ROW_MAJOR  101
#define LAPACK_COL_MAJOR  102

Error Codes

#define LAPACK_WORK_MEMORY_ERROR       -1010  /* Failed to allocate memory
                                            for a working array */
#define LAPACK_TRANSPOSE_MEMORY_ERROR  -1011  /* Failed to allocate memory
                                            for transposed matrix */

If the return value is -i, the -i-th parameter has an invalid value.

Function Prototypes

Some oneMKL functions differ in data types they support and vary in the parameters they take.

Each function type has a unique prototype defined. Use this prototype when you call the function from your application program. In most cases, oneMKL supports four distinct floating-point precisions. Each corresponding prototype looks similar, usually differing only in the data type. To avoid listing all the prototypes in every supported precision, a generic prototype template is provided. For example,

lapack_int LAPACKE_