3
votes

Intro: I've developed an application in C++ that utilizes LAPACK(LAPACKE) and MPI, all on Windows. Works OK in Windows (compiling and linking are taken care of through Code::Blocks IDE), but the execution is way too slow. Hence, I want to migrate the code to our small "supercomputer" running under CentOS Linux where we have already installed GNU C++, MPICH2 and LAPACK.

Questions: How to COMPILE/LINK and RUN a C++ code that calls LAPACKE on Linux/CentOS? Do I have to have GNU Fortran installed on the CentOS machine in order to compile/link/run C++ with LAPACK(LAPACKE)?

THANKS A LOT!!!

1

1 Answers

5
votes

I assume that my Debian is close enough to your CentOS for these tricks to work...

1) Check that LAPACKE is installed on your computed.

  • You can search locations like /usr/include, /usr/local/include for files lapacke.h, lapacke_config.h, lapacke_mangling.h, lapacke_mangling_with_flags.h and lapacke_utils.h.
  • You can search locations like /usr/lib or /usr/local/lib for the static library lapacke.a or the dynamic library lapacke.so or lapacke.so.3.

If these files are missing, consider installing the packages liblapacke and liblapacke-dev. Alternatively, (in particular if you don't have root privileges), you can download source of netlib's lapack and lapacke at http://www.netlib.org/lapack/#_lapack_version_3_6_1 To compile LAPACKE, rename make.inc.example to make.inc, then type:

    make
    make lapackelib

The include files will be located in lapack-3.6.1/LAPACKE/include and the library will be in lapack-3.6.1. gcc and gfortran are useful to recompile lapack and lapacke from scratch.

2) Let's compile a simple code based on this example:

#include <iostream>
#include <string>
#include <fstream>

#include <stdlib.h>
#include <stdio.h>
#include <math.h>

#include <lapacke.h>
#include "mpi.h"

void print_matrix_rowmajor(const char* desc, lapack_int m, lapack_int n, double* a, lapack_int lda );

int main(int argc, char *argv[])
{
    MPI_Init(&argc,&argv);
    std::cout << "Start..." << std::endl;
    //std::string fn_VALS;

    /* Locals */
    double A[5][3] = {1,1,1,2,3,4,3,5,2,4,2,5,5,4,3};
    double b[5][2] = {-10,-3,12,14,14,12,16,16,18,16};
    lapack_int info,m,n,lda,ldb,nrhs;

    /* Initialization */
    m = 5;
    n = 3;
    nrhs = 2;
    lda = 3;
    ldb = 2;

    /* Print Entry Matrix */
    print_matrix_rowmajor( "Entry Matrix A", m, n, *A, lda );
    /* Print Right Rand Side */
    print_matrix_rowmajor( "Right Hand Side b", n, nrhs, *b, ldb );
    printf( "\n" );
    /* Executable statements */
    printf( "LAPACKE_dgels (row-major, high-level) Example Program Results\n" );
    /* Solve least squares problem*/
    info = LAPACKE_dgels(LAPACK_ROW_MAJOR,'N',m,n,nrhs,*A,lda,*b,ldb);

    /* Print Solution */
    print_matrix_rowmajor( "Solution", n, nrhs, *b, ldb );
    printf( "\n" );


    std::cout << "info = " << info << std::endl;




    std::cout << "Done :-) !!!" <<std::endl;
    MPI_Finalize();
    return 0;
}


////////////////////////////////////////////////////////* Auxiliary routine: printing a matrix */
void print_matrix_rowmajor(const char* desc, lapack_int m, lapack_int n, double* a, lapack_int lda )
{
    lapack_int i, j;
    printf( "\n %s\n", desc );
    for( i = 0; i < m; i++ )
    {
        for( j = 0; j < n; j++ )
        {
            printf( " %6.2f", a[i*lda+j]);
        }
        printf( "\n" );
    }
}
//=======================================

The command to compile is:

mpiCC main.cpp -o main -llapacke -llapack -lblas -lm -Wall

If the include files are in a particular folder, use -I/usr/pathtolapackedoth. Similarly, if the library is in a particula folder, try -L/usr/lib/pathtoliblapackedota. Depending on how MPICH2 was installed, it is likely that mpiCC wraps g++. You can type mpiCC --version to learn more. To run it using 2 processes:

mpirun -np 2 main

Finally, you do not have to install GNU Fortran installed on the CentOS machine in order to compile/link/run C++ with LAPACK(LAPACKE). Indeed, it is only required if you wish to recompile LAPACK from scratch.