blas-lapack在windows下的ifort-makefile

THE LAPACKE C INTERFACE TO LAPACK

Contents

1. Introduction 1

1.1. Naming Schemes 1

1.2. Complex Types 2

1.3. Array Arguments 2

1.4. Aliasing of Arguments 2

1.5. INFO Parameters 2

1.6. NaN Checking 3

2.2. Complex Functions 4

2.3. Mixed Precision Functions 5

3. Examples 5

3.1. Calling DGEQRF 5

3.2. Calling CUNGQR 5

3.3. Calling DGELS 6

3.4. Calling CGEQRF and the CBLAS 7

1. Introduction

This document describes a two-level C interface to LAPACK, consisting

of a high-level interface and a middle-level interface. The high-level interface

to take the FORTRAN LAPACK routine name, make it lower case, and

add the prefix LAPACKE . For example, the LAPACK subroutine DGETRF

becomes LAPACKE dgetrf.

Date: September 30, 2011.

1

2 THE LAPACKE C INTERFACE TO LAPACK

The naming scheme for the middle-level interface is to take the FORTRAN

LAPACK routine name, make it lower case, then add the prefix LAPACKE

and the suffix work. For example, the LAPACK subroutine DGETRF be-

comes LAPACKE dgetrf work.

defined type, C++ STL complex types, or a custom complex type. See

lapacke.h for more details.

1.3. Array Arguments. Arrays are passed as pointers, not as a pointer to

pointers. All the LAPACKE routines that take one or more 2D arrays as a

pointer receive a single extra parameter of type int. This argument must be

equal to either LAPACK ROW MAJOR or LAPACK COL MAJOR which are

defined in lapacke.h, specifying whether the arrays are stored in row-major

or column-major order. If a routine has multiple array inputs, they must all

use the same ordering.

Note that using row-major ordering may require more memory and time

from one row to the next are assumed to be a leading dimension apart.

For column-major 2D arrays, elements within a column are assumed to be

contiguous and elements from one column to the next are assumed to be a

leading dimension apart.

1.4. Aliasing of Arguments. Unless specified otherwise, only input argu-

ments (that is, scalars passed by values and arrays specified with the const

qualifier) may be legally aliased on a call to C interface to LAPACK.

1.5. INFO Parameters. The LAPACKE interface functions set their

lapack int return value to the value of the INFO parameter, which contains

information such as error and exit conditions. This differs from LAPACK

the value is not adjusted for zero-based indexing.

THE LAPACKE C INTERFACE TO LAPACK 3

1.6. NaN Checking. The high-level interface includes an optional, on by

default, NaN check on all matrix inputs before calling any LAPACK routine.

This option affects all routines. If the inputs contain any NaNs, the input

parameter corresponding matrix will be flagged with an INFO parameter

ing LAPACK DISABLE NAN CHECK macro in lapacke.h. The middle-level

interface does not contain the NaN check.

1.7. Integers. Variables with the FORTRAN type integer are converted to

lapack int in LAPACKE. This conforms with modifiable integer type size,

especially given ILP64 programming model: re-defining lapack int as long

int (8 bytes) will be enough to support this model, as lapack int is defined

as int (4 bytes) by default, supporting LP64 programming model.

1.8. Logicals. FORTRAN logicals are converted to lapack logical, which is

defined as lapack int.

1.9. Memory Management. All memory management is handled by the

arrays, error notification is needed in the event of a user running out of

memory.

If a work array cannot be allocated, LAPACK WORK MEMORY ERROR

is returned by the function; if there was insufficient memory to complete a

transposition, LAPACK TRANSPOSE MEMORY ERROR is returned.

2. Function List

This section list the currently available LAPACK subroutines that are

available in the LAPACKE C interface. The LAPACK base names are given

below; the corresponding LAPACKE function name is LAPACKE xbase or

LAPACKE xbase work where x is the type: s or d for single or double preci-

their parameters.