example_mdof1.c File Reference

Go to the source code of this file.

Detailed Description

Example to show how to use the SPM library with a variadic 0-based multi-dof sparse matrix allocated through the library but initialized by the user.

Copyright

2020-2024 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux. All rights reserved.

Version

1.2.4

Author

Mathieu Faverge

Tony Delarue

Date

2024-06-26

   /
#include <spm.h>
 
/**
   @brief Create an IJV symmetric laplacian structure matrix for a regular cube of
   dimensions (dim1 x dim2 x dim3) and with 1-based values as in Fortran arrays.
  /
static inline void
spm_example_fill_ijv_struct( spmatrix_t *spm,
                             spm_int_t   dim1,
                             spm_int_t   dim2,
                             spm_int_t   dim3 )
{
    spm_int_t i, j, k, d, n, nnz;
    spm_int_t mdim1, mdim2, mdim3;
 
    mdim1 = dim1 - 1;
    mdim2 = dim2 - 1;
    mdim3 = dim3 - 1;
    n   = 0;
    nnz = 0;
    d   = 0;
    for( i = 0; i < dim1; i++ )
    {
        for( j = 0; j < dim2; j++ )
        {
            for( k = 0; k < dim3; k++ )
            {
                /*
                   Start with the diagonal element connected with its two
                   neigbours in the three directions
                  /
                spm->rowptr[nnz] = i + dim1 * j + dim1 * dim2 * k;
                spm->colptr[nnz] = i + dim1 * j + dim1 * dim2 * k;
                nnz++;
 
                spm->dofs[n] = d;
                n++;
 
                /* Number of degrees of freedom of the current element */
                d += (( i + j + k ) % 3) + 1;
 
                /* Add connexions */
                if ( i < mdim1 ) {
                    spm->rowptr[nnz] = (i+1) + dim1 * j + dim1 * dim2 * k;
                    spm->colptr[nnz] =  i    + dim1 * j + dim1 * dim2 * k;
                    nnz++;
                }
 
                if ( j < mdim2 ) {
                    spm->rowptr[nnz] = i + dim1 * (j+1) + dim1 * dim2 * k;
                    spm->colptr[nnz] = i + dim1 *  j    + dim1 * dim2 * k;
                    nnz++;
                }
 
                if ( k < mdim3 ) {
                    spm->rowptr[nnz] = i + dim1 * j + dim1 * dim2 * (k+1);
                    spm->colptr[nnz] = i + dim1 * j + dim1 * dim2 *  k;
                    nnz++;
                }
            }
        }
    }
    /* Final value */
    spm->dofs[n] = d;
 
    assert( nnz == ((2*(dim1)-1) * dim2 * dim3 + (dim2-1)*dim1*dim3 + dim2*dim1*(dim3-1)) );
}
 
/**
   @brief Fill-in the values array of a variadic dof IJV sparse matrix.
  /
static inline void
spm_example_fill_ijv_values( spmatrix_t *spm,
                             spm_int_t   dim1,
                             spm_int_t   dim2,
                             spm_int_t   dim3 )
{
    spm_int_t v, l, i, j, k, m, n;
    spm_int_t mdim1, mdim2, mdim3;
    spm_int_t row, col, dofrow, dofcol;
    double   *values = spm->values;
    double    alpha = 8.;
    double    beta  = 0.5;
    double    value;
 
    mdim1 = dim1 - 1;
    mdim2 = dim2 - 1;
    mdim3 = dim3 - 1;
    v = 0;
    for( l=0; l<spm->nnz; l++ )
    {
        /* Get indices in the mesh */
        k =  l / ( dim1 * dim2 );
        j = (l % ( dim1 * dim2 )) / dim1;
        i =  l % dim1;
 
        col = spm->colptr[l];
        row = spm->rowptr[l];
        dofcol = spm->dofs[ col+1 ] - spm->dofs[ col ];
        dofrow = spm->dofs[ row+1 ] - spm->dofs[ row ];
 
        /* diagonal element */
        if ( col == row ) {
            value = 6.;
 
            /*
               Handle limit cases
              /
            if ( i == 0     ) { value -= 1.; }
            if ( i == mdim1 ) { value -= 1.; }
            if ( j == 0     ) { value -= 1.; }
            if ( j == mdim2 ) { value -= 1.; }
            if ( k == 0     ) { value -= 1.; }
            if ( k == mdim3 ) { value -= 1.; }
 
            value *= alpha;
        }
        else {
            value = - beta;
        }
 
        /* Fill the element matrix */
        for ( n=0; n<dofcol; n++ ) {
            for ( m=0; m<dofrow; m++ ) {
                values[v] = value;
                v++;
            }
        }
    }
 
    assert( v == spm->nnzexp );
}
 
/**
   @brief Create a laplacian manually
  /
void
spm_example_create_laplacian( spmatrix_t *spm )
{
    spm_int_t dim1, dim2, dim3;
    dim1 = 6;
    dim2 = 6;
    dim3 = 6;
 
    /*
       Initialize all non computed fields.
      /
    spmInit( spm );
 
    /*
       Set the fields according to what we want to generate.
       All non computed fields must be set.
      /
    spm->baseval = 0;
    spm->mtxtype = SpmSymmetric;
    spm->flttype = SpmPattern;
    spm->fmttype = SpmIJV;
    spm->n       = dim1 * dim2 * dim3;
    spm->nnz     = dim1 * dim2 * dim3;       /* Diagonal elements           */
    spm->nnz    += (dim1 - 1) * dim2 * dim3; /* Connexions along first axe  */
    spm->nnz    += dim1 * (dim2 - 1) * dim3; /* Connexions along second axe */
    spm->nnz    += dim1 * dim2 * (dim3 - 1); /* Connexions along third axe  */
 
    /* The full matrix is replicated on all nodes */
    spm->replicated = 1;
 
    /* Force gN to allocate dof array in spmAlloc() */
    spm->gN  = spm->n;
    spm->dof = -1;
 
    /*
       Let's allocate first only the structure arrays
       That is why we set flttype to SpmPattern
      /
    spmAlloc( spm );
 
    /*
       Fill the structure arrays with the generator.
      /
    spm_example_fill_ijv_struct( spm, dim1, dim2, dim3 );
 
    /*
       Update the computed fields now that the dofs arrays is initialized.
      /
    spmUpdateComputedFields( spm );
 
    /*
       Switch to the arithmetic type of interest and allocate the value array
      /
    spm->layout  = SpmColMajor; /* Storage of the element matrices */
    spm->flttype = SpmDouble;
    spmAlloc( spm );
 
    /*
       Fill the value array with the generator.
      /
    spm_example_fill_ijv_values( spm, dim1, dim2, dim3 );
}
 
int main( int argc, char **argv )
{
    spmatrix_t spm;
    spm_int_t  nrhs  = 10;
    double     alpha = 2.;
    spm_int_t  size, ldb, ldx;
    double     epsilon, norm;
    void      *x, *b;
    int        rc = 0;
 
    /*
       MPI need to be initialize before any call to spm library if it has been
       compiled wth MPI support.
      /
#if defined(SPM_WITH_MPI)
    MPI_Init( &argc, &argv );
#endif
 
    /*
       Create a user made sparse matrix.
      /
    spm_example_create_laplacian( &spm );
 
    /*
       Print basic information about the sparse matrix
      /
    spmPrintInfo( &spm, stdout );
 
    /*
       Possibly convert the SPM into another format (CSC may be needed for a few routines)
      /
    spmConvert( SpmCSC, &spm );
 
    /*
       Compute the frobenius norm of the spm
      /
    norm = spmNorm( SpmFrobeniusNorm, &spm );
    fprintf( stdout, "  || A ||_f:    %e\n", norm );
 
    /*
       Scale the sparse matrix.
      /
    if ( norm > 0. ) {
        spmScal( 1. / norm, &spm );
    }
 
    /*
       Create a random matrix x to test products (multiple right hand side).
       Note that you can get the size to allocate with spm_size_of() that
       returns the size of each value.
      /
    ldb  = spm.nexp > 1 ? spm.nexp : 1;
    ldx  = spm.nexp > 1 ? spm.nexp : 1;
    size = spm_size_of( spm.flttype ) * spm.nexp * nrhs;
    x    = malloc( size );
    rc   = spmGenMat( SpmRhsRndX, nrhs, &spm, &alpha, 24356, x, ldx );
    if ( rc != SPM_SUCCESS ) {
        free( x );
        spmExit( &spm );
        return rc;
    }
 
    /*
       Compute b = A * x
      /
    b = malloc( size );
    spmMatMat( SpmNoTrans, nrhs, 1., &spm, x, ldx, 0., b, ldb );
 
    /*
       The following routines helps to check results obtained out of solvers by
       computing b = b - A * x with a given precision.
      /
    if ( (spm.flttype == SpmDouble   ) ||
         (spm.flttype == SpmComplex64) )
    {
        epsilon = 1e-15;
    }
    else {
        epsilon = 1e-7;
    }
    epsilon = epsilon * spm.nnzexp / spm.gN;
    rc = spmCheckAxb( epsilon, nrhs, &spm, NULL, 1, b, ldb, x, ldx );
 
    free( x );
    free( b );
    spmExit( &spm );
 
#if defined(SPM_WITH_MPI)
    MPI_Finalize();
#endif
 
    (void)argc;
    (void)argv;
    return rc;
}
/**

Definition in file example_mdof1.c.