CLAPACK_SC is a C++ program which illustrates the use of CLAPACK, a C translation of the FORTRAN77 BLAS and LAPACK linear algebra libraries, as installed at the FSU Department of Scientific Computing.
Get a copy of the one page instruction sheet.
By translating the LAPACK library into C, several important linear algebra operations were made available in efficient, reliable versions, including:
The translation was done using the automatic F2C translator. As a consequence, the resulting C source code is at best unpleasant to read. Moreover, the user's C code must be written and processed in a particular way.
First, the user code must have the necessary "include" statements. This means adding the statement:
# include "clapack.h"(Normally, three include statements are needed, but the FSU SC version has simplified this interface.)
Secondly, all variables that will be passed to a CLAPACK function must be declared using types that can be handled by the FORTRAN package. In general, this only affects integer variables; as a rule this means that if PIV is an integer scalar, vector or array that will be passed to CLAPACK, its type must be either "integer" (a type defined by f2c.h), or else as "long int" (a standard C type). Declaring such a variable as "int" will not work!
Each user accessible routine in the FORTRAN version of LAPACK has a corresponding CLAPACK version. However, to access the CLAPACK version, the user must specify the name of the routine in lower case letters only, and must append an underscore to the name. Thus, to access a LAPACK routine such as DGETRF(), the user's C code must look something like this:
dgetrf_ ( list of arguments )
Because all FORTRAN subroutine arguments can be modified during the execution of the subroutine, the CLAPACK interface requires that every argument in the argument list must be modifiable. In cases where a vector or array is being passed, this happens automatically. However, when passing a scalar variable, such as "N", the size of the linear system, or "LDA", the leading dimension of array A, or "INFO", an error return flag, it is necessary to prefix the name with an ampersand. Thus, a call to DGESV() might look like:
dgesv_ ( &n, &nrhs, a, &lda, ipiv, b, &ldb, &info );because n, nrhs, lda, ldb and info are scalar variables.
The vector and matrix arguments to the CLAPACK routine don't require the ampersand. Moreover, vectors (singly indexed lists of numbers) essentially work the same in C and FORTRAN, so it's not difficult to correctly set up vector arguments for CLAPACK. However, matrices (doubly indexed sets of data) are handled differently, and the user's C code must either set up the data in a FORTRAN way immediately, or else set it up in a way natural to C and then convert the data to make a FORTRAN copy.
Let's assume that we have an M by N set of data, and to be concrete, let's consider an example where M = 3 and N = 2. In C, it would be natural to declare this data as follows:
double a[3][2] = { { 11, 12 }, { 21, 22 }, { 31, 32 } };In this case, the (I,J) entry (using 0-based indexing) can be retrieved as a[i][j].
However, FORTRAN essentially stores a matrix as a vector, in which the data is stored on column at a time. Thus, if we wished to pass the example data to CLAPACK as an array, we might instead use the following declaration:
double b[3*2] = { 11, 21, 31, 12, 22, 32 };In this case, the (I,J) entry (using 0-based indexing) can be retrieved as b[i+j*3] where the number 3 is the "leading dimension" of the array, that is, the length of one column.
But suppose we need to build the array using the double indexed version, although we know we have to pass a single indexed copy to CLAPACK? Then we can start with the following declarations:
double a[3][2]; double b[3*2];and calculate the entries of a using double indexing, and then copy the information into b using code like the following:
for ( j = 0; j < 2; j++ ) { for ( i = 0; i < 3; i++ ) { b[i+j*3] = a[i][j]; } }after which, the vector b will contain a copy of the data that is in a, suitable for use by CLAPACK.
Both LAPACK and CLAPACK allow you to store M by N data inside arrays of larger size. In that case, the actual first dimension of the double dimensioned array is called the "leading dimension", and its value must be available whenever entries of the smaller array must be located inside the bigger array. I will assume for now that you always make your arrays exactly big enough to hold the data you are interested in. Typically, this will mean that the variables LDA and LDB are equal to N.
Once you think you've got your user code all set up, you need to compile and load your program. Compilation requires access to a copy of the CLAPACK include file, and that depends on how CLAPACK was installed on your system. For our local system, the include file is stored in /usr/common/clapack, so the compile statement might be
g++ -I/usr/common/clapack myprog.cpp
The load procedure requires access to the CLAPACK library. For our local system, the library is stored as libclapack.a in /usr/local/common, and so the load statement is:
g++ myprog.o -L/usr/common/clapack -lclapack -lmwhich should create an executable program.
The source code for the CLAPACK library is available from http://www.netlib.org/clapack . Actually setting up the library for use can be surprisingly awkward and failure-prone.
The computer code and data files described and made available on this web page are distributed under the GNU LGPL license.
This refers to the EXAMPLES presented here. The CLAPACK library itself is licensed under a different arrangement.
The examples for CLAPACK are available in a C version and a C++ version.
BLAS1_C, a C++ library which contains basic linear algebra subprograms (BLAS) for vector-vector operations, using single precision complex arithmetic, by Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh.
BLAS1_D, a C++ library which contains basic linear algebra subprograms (BLAS) for vector-vector operations, using double precision real arithmetic, by Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh.
BLAS1_S, a C++ library which contains basic linear algebra subprograms (BLAS) for vector-vector operations, using single precision real arithmetic, by Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh.
BLAS1_Z, a C++ library which contains basic linear algebra subprograms (BLAS) for vector-vector operations, using double precision complex arithmetic, by Charles Lawson, Richard Hanson, David Kincaid, Fred Krogh.
CLAPACK, C++ programs which illustrate the use of the CLAPACK library, a C translation of the FORTRAN77 BLAS and LAPACK linear algebra libraries, including single and double precision, real and complex arithmetic.
LINPACK_D, a C++ library which solves linear systems using double precision real arithmetic;
LINPACK_S, a C++ library which solves linear systems using single precision real arithmetic;
CLAPACK_PRB is a complicated example that uses vectors to store the matrix information.
CLAPACK_PRB2 is a simplified example that stores the matrix as a double-indexed array, typical for C programs, but then shows how to convert it to the form that CLAPACK needs.
CLAPACK_PRB3 shows how to use DGBTRF and DGBTRS to factor a banded matrix and solve a related linear system. The banded matrix is stored as a double-indexed array of bands, but then the program converts the matrix data to the form that CLAPACK needs.
You can go up one level to the C++ source codes.