step-by-step.c File Reference

A step-by-step example that runs one full analyze (ordering, symbolic factorization, analyze), then loops over 2 factorizations that are both used for 2 solves each. More...

Go to the source code of this file.

Detailed Description

A step-by-step example that runs one full analyze (ordering, symbolic factorization, analyze), then loops over 2 factorizations that are both used for 2 solves each.

Copyright

2015-2021 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux. All rights reserved.

Version

6.2.0

Author

Mathieu Faverge

Pierre Ramet

Tony Delarue

Date

2021-04-07

  /
#include <pastix.h>
#include <spm.h>
 
int main (int argc, char **argv)
{
    pastix_data_t  *pastix_data = NULL; /*< Pointer to the storage structure required by pastix */
    pastix_int_t    iparm[IPARM_SIZE];  /*< Integer in/out parameters for pastix                */
    double          dparm[DPARM_SIZE];  /*< Floating in/out parameters for pastix               */
    spm_driver_t    driver;
    char           *filename = NULL;
    spmatrix_t     *spm, spm2;
    void           *x, *b, *x0 = NULL;
    size_t          size;
    int             check = 1;
    int             nrhs  = 10;
    int             rc    = 0;
    int             nfact = 2;
    int             nsolv = 2;
    long            i,j;
 
    /**
     * Initialize parameters to default values
     */
    pastixInitParam( iparm, dparm );
 
    /**
     * Get options from command line
     */
    pastixGetOptions( argc, argv,
                      iparm, dparm,
                      &check, &driver, &filename );
 
    /**
     * Startup PaStiX
     */
    pastixInit( &pastix_data, MPI_COMM_WORLD, iparm, dparm );
 
    /**
     * Read the sparse matrix with the driver
     */
    spm = malloc( sizeof( spmatrix_t ) );
    rc = spmReadDriver( driver, filename, spm );
    free( filename );
    if ( rc != SPM_SUCCESS ) {
        pastixFinalize( &pastix_data );
        return rc;
    }
 
    spmPrintInfo( spm, stdout );
 
    rc = spmCheckAndCorrect( spm, &spm2 );
    if ( rc != 0 ) {
        spmExit( spm );
        *spm = spm2;
        rc = 0;
    }
 
    /**
     * Generate a Fake values array if needed for the numerical part
     */
    if ( spm->flttype == SpmPattern ) {
        spmGenFakeValues( spm );
    }
 
    /**
     * Perform ordering, symbolic factorization, and analyze steps
     */
    pastix_subtask_order( pastix_data, spm, NULL );
    pastix_subtask_symbfact( pastix_data );
    pastix_subtask_reordering( pastix_data );
    pastix_subtask_blend( pastix_data );
 
    /**
     * Normalize A matrix (optional, but recommended for low-rank functionality)
     */
    double normA = spmNorm( SpmFrobeniusNorm, spm );
    spmScal( 1./normA, spm );
 
    size = pastix_size_of( spm->flttype ) * spm->nexp * nrhs;
    x = malloc( size );
    b = malloc( size );
    if ( check > 1 ) {
        x0 = malloc( size );
    }
 
    /* Do nfact factorization */
    for (i = 0; i < nfact; i++)
    {
        /**
         * Perform the numerical factorization
         */
        pastix_subtask_spm2bcsc( pastix_data, spm );
        pastix_subtask_bcsc2ctab( pastix_data );
        pastix_subtask_sopalin( pastix_data );
        for (j = 0; j < nsolv; j++)
        {
            /**
             * Generates the b and x vector such that A * x = b
             * Compute the norms of the initial vectors if checking purpose.
             */
            if ( check )
            {
                spmGenRHS( SpmRhsRndX, nrhs, spm, x0, spm->nexp, b, spm->nexp );
                memcpy( x, b, size );
            }
            else {
                spmGenRHS( SpmRhsRndB, nrhs, spm, NULL, spm->nexp, x, spm->nexp );
 
                /* Apply also normalization to b vectors */
                spmScalMat( 1./normA, spm, nrhs, b, spm->nexp );
 
                /* Save b for refinement */
                memcpy( b, x, size );
            }
 
            /**
             * Solve the linear system
             */
            pastix_task_solve( pastix_data, nrhs, x, spm->nexp );
            pastix_task_refine( pastix_data, spm->nexp, nrhs, b, spm->nexp, x, spm->nexp );
 
            if ( check ) {
                rc |= spmCheckAxb( dparm[DPARM_EPSILON_REFINEMENT], nrhs, spm, x0, spm->nexp, b, spm->nexp, x, spm->nexp );
            }
        }
    }
 
    spmExit( spm );
    free( spm );
    free( b );
    free( x );
    if ( x0 ) {
        free( x0 );
    }
    pastixFinalize( &pastix_data );
 
    return rc;
}
 
/**
 * 

Definition in file step-by-step.c.