codelet_dcmg_diag.c

Go to the documentation of this file.

#include "../include/starpu_exageostat.h"
#define MIN(X, Y) (((X) < (Y)) ? (X) : (Y))
#define MAX(X, Y) (((X) > (Y)) ? (X) : (Y))

static void CORE_dcmg_starpu(void *buffers[],void *cl_arg){
        int m, n, m0, n0;
        location * l1;
        location * l2;
        double * theta;
        double * A;
    int distance_metric;
        theta   = (double *) malloc(3* sizeof(double));
        A   = (double *)STARPU_MATRIX_GET_PTR(buffers[0]);

        starpu_codelet_unpack_args(cl_arg, &m, &n, &m0, &n0, &l1, &l2, &theta[0], &theta[1], &theta[2], &distance_metric);


        core_dcmg(A, m, n, m0, n0, l1, l2, theta, distance_metric);

}

static struct starpu_codelet cl_dcmg =
{
                .where      = STARPU_CPU,
                .cpu_funcs  = {CORE_dcmg_starpu},
                .nbuffers   = 1,
                .modes      = {STARPU_W},
        .name       = "dcmg"
};

/***************************************************************************/
int MORSE_MLE_dcmg_diag_Tile_Async(MORSE_enum uplo, MORSE_desc_t *descA, MORSE_sequence_t *sequence, MORSE_request_t  *request, location *l1, location *l2, double *theta, char *dm, int diag_thick) {

        MORSE_context_t *morse;
        MORSE_option_t options;
        morse = morse_context_self();

      
  if (sequence->status != MORSE_SUCCESS)
                return -2;
        RUNTIME_options_init(&options, morse, sequence, request);


        int m, n, m0, n0;
    int distance_metric = strcmp(dm,"gc") == 0? 1 : 0 ; 
        int tempmm, tempnn;
        MORSE_desc_t A = *descA;
        struct starpu_codelet *cl = &cl_dcmg;

    int k = 0;

        for(m = 0; m < A.mt;)
        {
        int v = m;
        
        for (k=0 ; k<diag_thick && m < A.mt;k++)
        {   
                    tempmm = m == A.mt -1 ? A.m- m* A.mb : A.mb;
               
         
                for (n = v; n < A.nt && n < v+diag_thick; n++) {
                    //Escape unwanted tiles 
    
                    //printf ("%d - %d\n", m , n);
                                tempnn = n == A.nt -1 ? A.n - n * A.nb : A.nb;
                                m0= m * A.mb;
                                n0= n * A.nb;
                                starpu_insert_task(starpu_mpi_codelet(cl),
                                            STARPU_VALUE, &tempmm, sizeof(int),
                                            STARPU_VALUE, &tempnn, sizeof(int),
                                            STARPU_VALUE, &m0, sizeof(int),
                                            STARPU_VALUE, &n0, sizeof(int),
                                            STARPU_W, RTBLKADDR(descA, sizeof(double)*ldam*tempnn, m, n),
                                            STARPU_VALUE, &l1, sizeof(location*),
                                            STARPU_VALUE, &l2, sizeof(location*),
                                            STARPU_VALUE, &theta[0], sizeof(double),
                                            STARPU_VALUE, &theta[1], sizeof(double),
                                            STARPU_VALUE, &theta[2], sizeof(double),
                                            STARPU_VALUE, &distance_metric, sizeof(int),                         
                        0);
                }
                m++;
                }

        }

    //RUNTIME_data_flush( sequence, descA );
    //MORSE_TASK_flush_desc( &options, MorseUpperLower, descA);
        RUNTIME_options_ws_free(&options);
        RUNTIME_options_finalize(&options, morse);
        //MORSE_TASK_dataflush_all();        
    return MORSE_SUCCESS;
}