static double
RunTest(real_Double_t *t_, struct user_parameters* params)
{
double t;
int64_t N = params->matrix_size;
int64_t NB = params->blocksize;
int check = params->check;
int uplo = PlasmaUpper;
double check_res = 0;
/* Allocate Data */
PLASMA_desc *descA = NULL;
double* ptr = malloc(N * N * sizeof(double));
PLASMA_Desc_Create(&descA, ptr, PlasmaRealDouble, NB, NB, NB*NB, N, N, 0, 0, N, N);
#pragma omp parallel
{
#pragma omp single
{
plasma_pdplgsy_quark( (double)N, *descA, 51 );
}
}
/* Save A for check */
double *A = NULL;
if(check) {
A = (double*)malloc(N * N * sizeof(double));
plasma_pdtile_to_lapack_quark(*descA, (void*)A, N);
}
/* PLASMA DPOSV */
START_TIMING();
#pragma omp parallel
{
#pragma omp single
{
plasma_pdpotrf_quark(uplo, *descA);
}
}
STOP_TIMING();
/* Check the solution */
if ( check )
{
PLASMA_desc *descB = NULL;
double* ptr = (double*)malloc(N * sizeof(double));
PLASMA_Desc_Create(&descB, ptr, PlasmaRealDouble, NB, NB, NB*NB, N, 1, 0, 0, N, 1);
plasma_pdpltmg_seq(* descB, 7672 );
double* B = (double*)malloc(N * sizeof(double));
plasma_pdtile_to_lapack_quark(*descB, (void*)B, N);
PLASMA_dpotrs_Tile( uplo, descA, descB );
double* X = (double*)malloc(N * sizeof(double));
plasma_pdtile_to_lapack_quark(*descB, (void*)X, N);
check_res = d_check_solution(N, N, 1, A, N, B, X, N);
PASTE_CODE_FREE_MATRIX( descB );
free( A );
free( B );
free( X );
}
PASTE_CODE_FREE_MATRIX( descA );
return check_res;
}
void PLASMA_DPOTRS_TILE(PLASMA_enum *uplo, intptr_t *A, intptr_t *B, int *INFO)
{ *INFO = PLASMA_dpotrs_Tile(*uplo, (PLASMA_desc *)(*A), (PLASMA_desc *)(*B)); }
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
double *A = NULL, *AT, *b = NULL, *bT, *x;
real_Double_t t;
PLASMA_desc *descA, *descB;
int nb, nb2, nt;
int n = iparam[TIMING_N];
int nrhs = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = n;
int ldb = n;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
#if defined(PLASMA_CUDA)
core_cublas_init();
#endif
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* } */
nb = iparam[TIMING_NB];
nb2 = nb * nb;
nt = n / nb + ((n % nb == 0) ? 0 : 1);
/* Allocate Data */
#if defined(PLASMA_CUDA)
cudaHostAlloc((void**)&AT, nt*nt*nb2*sizeof(double), cudaHostAllocPortable);
#else
AT = (double *)malloc(nt*nt*nb2*sizeof(double));
#endif
/* Check if unable to allocate memory */
if ( !AT ){
printf("Out of Memory \n ");
exit(0);
}
// cudaHostRegister(AT, nt*nt*nb2*sizeof(double), cudaHostRegisterPortable);
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaRealDouble, nb, nb, nb*nb, n, n, 0, 0, n, n);
PLASMA_dplgsy_Tile((double)n, descA, 51 );
/* Save AT in lapack layout for check */
if ( check ) {
A = (double *)malloc(lda*n *sizeof(double));
PLASMA_Tile_to_Lapack(descA, (void*)A, n);
}
/* PLASMA DPOSV */
t = -cWtime();
PLASMA_dpotrf_Tile(PlasmaUpper, descA);
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
b = (double *)malloc(ldb*nrhs *sizeof(double));
bT = (double *)malloc(nt*nb2 *sizeof(double));
x = (double *)malloc(ldb*nrhs *sizeof(double));
LAPACKE_dlarnv_work(1, ISEED, nt*nb2, bT);
PLASMA_Desc_Create(&descB, bT, PlasmaRealDouble, nb, nb, nb*nb, n, nrhs, 0, 0, n, nrhs);
PLASMA_Tile_to_Lapack(descB, (void*)b, n);
PLASMA_dpotrs_Tile( PlasmaUpper, descA, descB );
PLASMA_Tile_to_Lapack(descB, (void*)x, n);
dparam[TIMING_RES] = d_check_solution(n, n, nrhs, A, lda, b, x, ldb,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
PLASMA_Desc_Destroy(&descB);
free( A );
free( b );
free( bT );
free( x );
}
PLASMA_Desc_Destroy(&descA);
PLASMA_Finalize();
#if defined(PLASMA_CUDA)
cudaFreeHost(AT);
#else
free(AT);
#endif
// cudaHostUnregister(AT);
return 0;
}
