int plasma_init() {
#ifdef HIPLAR_WITH_PLASMA
int info;
R_PLASMA_NUM_THREADS = 1;
if (getenv("R_PLASMA_NUM_THREADS") != NULL) {
R_PLASMA_NUM_THREADS = atoi(getenv("R_PLASMA_NUM_THREADS"));
} else {
Rprintf("The envirnment variable R_PLASMA_NUM_THREADS is not set.\n");
Rprintf("Using one thread for PLASMA.\nPlease set R_PLASMA_NUM_THREADS to the number of actual cores.\n\n");
//printf("\nThe envirnment variable R_PLASMA_NUM_THREADS is not set.\n");
//printf("Using one thread for PLASMA.\nPlease set R_PLASMA_NUM_THREADS to the number of actual cores.\n\n");v
}
/* Init PLASMA */
info = PLASMA_Init(R_PLASMA_NUM_THREADS);
if ((getenv("R_PLASMA_SCHED") != NULL) &&
(strcmp(getenv("R_PLASMA_SCHED"), "STATIC") == 0)) {
PLASMA_Set( PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
R_PLASMA_SCHED = 0;
} else {
PLASMA_Set( PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
R_PLASMA_SCHED = 1;
}
PLASMA_Version(&R_PLASMA_MAJOR, &R_PLASMA_MINOR, &R_PLASMA_MICRO);
if ((PLASMA_VERSION_MAJOR != R_PLASMA_MAJOR) ||
(PLASMA_VERSION_MINOR != R_PLASMA_MINOR) ||
(PLASMA_VERSION_MICRO != R_PLASMA_MICRO)) {
Rprintf("ERROR: PLASMA version mismatch\n");
Rprintf("ERROR: PLASMA version mismatch %d.%d.%d %d.%d.%d\n",
PLASMA_VERSION_MAJOR, PLASMA_VERSION_MINOR, PLASMA_VERSION_MICRO,
R_PLASMA_MAJOR, R_PLASMA_MINOR, R_PLASMA_MICRO);
}
return(info);
#endif
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
float *A, *Acpy = NULL, *b, *x;
real_Double_t t;
int *piv;
int m = iparam[TIMING_M];
int n = iparam[TIMING_N];
int nrhs = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = m;
int ldb = m;
/* Allocate Data */
A = (float *)malloc(lda*n*sizeof(float));
piv = (int *)malloc( min(m, n) * sizeof(int));
/* Check if unable to allocate memory */
if ( !A || !piv ){
printf("Out of Memory \n ");
return -1;
}
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
/* Initialize Data */
/*LAPACKE_slarnv_work(1, ISEED, n*lda, A);*/
PLASMA_splrnt(m, n, A, lda, 3456);
/* Save AT in lapack layout for check */
if ( check && (m == n) ) {
Acpy = (float *)malloc(lda*n*sizeof(float));
LAPACKE_slacpy_work(LAPACK_COL_MAJOR, 'A', m, n, A, lda, Acpy, lda);
}
t = -cWtime();
PLASMA_sgetrf( m, n, A, lda, piv );
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check && (m == n) )
{
b = (float *)malloc(ldb*nrhs *sizeof(float));
x = (float *)malloc(ldb*nrhs *sizeof(float));
LAPACKE_slarnv_work(1, ISEED, ldb*nrhs, x);
LAPACKE_slacpy_work(LAPACK_COL_MAJOR, 'A', n, nrhs, x, ldb, b, ldb);
PLASMA_sgetrs( PlasmaNoTrans, n, nrhs, A, lda, piv, x, ldb );
dparam[TIMING_RES] = s_check_solution(m, n, nrhs, Acpy, lda, b, x, ldb,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free( Acpy ); free( b ); free( x );
}
free( A );
free( piv );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
float *AT, *Q = NULL;
float *W;
PLASMA_desc *descA = NULL;
PLASMA_desc *descQ = NULL;
PLASMA_desc *descT = NULL;
real_Double_t t;
int nb, nb2, nt;
int n = iparam[TIMING_N];
int check = iparam[TIMING_CHECK];
int lda = n;
int uplo = PlasmaUpper;
int vec = PlasmaNoVec;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
nb = iparam[TIMING_NB];
nb2 = nb * nb;
nt = n / nb + ((n % nb == 0) ? 0 : 1);
/* Allocate Data */
AT = (float *)malloc(lda*n*sizeof(float));
W = (float *)malloc(n*sizeof(float));
if (vec == PlasmaVec){
Q = (float *)malloc(lda*n*sizeof(float));
if ( (!Q) ) {
printf("Out of Memory -Q-\n ");
return -2;
}
}
/* Check if unable to allocate memory */
if ( (!AT) || (!W) ) {
printf("Out of Memory -\n ");
return -2;
}
/* Initialize Data */
PLASMA_Desc_Create(&descA, AT, PlasmaRealFloat, nb, nb, nb*nb, lda, n, 0, 0, n, n);
PLASMA_splgsy_Tile((float)0.0, descA, 51 );
if (vec == PlasmaVec)
PLASMA_Desc_Create(&descQ, Q, PlasmaRealFloat, nb, nb, nb*nb, lda, n, 0, 0, n, n);
/* Save AT and bT in lapack layout for check */
if ( check ) {
}
/* Allocate Workspace */
PLASMA_Alloc_Workspace_ssyev(n, n, &descT);
t = -cWtime();
PLASMA_ssyev_Tile( vec, uplo, descA, W, descT, descQ );
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
}
/* DeAllocate Workspace */
PLASMA_Dealloc_Handle_Tile(&descT);
PLASMA_Desc_Destroy(&descA);
if (vec == PlasmaVec) {
PLASMA_Desc_Destroy(&descQ);
free( Q );
}
free( AT );
free( W );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
PLASMA_Complex64_t *A, *Acpy = NULL, *b = NULL, *x;
real_Double_t t;
int n = iparam[TIMING_N];
int nrhs = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = n;
int ldb = n;
/* Allocate Data */
A = (PLASMA_Complex64_t *)malloc(lda*n* sizeof(PLASMA_Complex64_t));
x = (PLASMA_Complex64_t *)malloc(ldb*nrhs*sizeof(PLASMA_Complex64_t));
/* Check if unable to allocate memory */
if ( (!A) || (!x) ) {
printf("Out of Memory \n ");
exit(0);
}
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
/* } */
/* Initialiaze Data */
PLASMA_zplghe((double)n, n, A, lda, 51 );
LAPACKE_zlarnv_work(1, ISEED, n*nrhs, x);
/* Save A and b */
if (check) {
Acpy = (PLASMA_Complex64_t *)malloc(lda*n* sizeof(PLASMA_Complex64_t));
b = (PLASMA_Complex64_t *)malloc(ldb*nrhs*sizeof(PLASMA_Complex64_t));
LAPACKE_zlacpy_work(LAPACK_COL_MAJOR,' ', n, n, A, lda, Acpy, lda);
LAPACKE_zlacpy_work(LAPACK_COL_MAJOR,' ', n, nrhs, x, ldb, b, ldb);
}
/* PLASMA ZPOSV */
t = -cWtime();
PLASMA_zposv(PlasmaUpper, n, nrhs, A, lda, x, ldb);
t += cWtime();
*t_ = t;
/* Check the solution */
if (check)
{
dparam[TIMING_RES] = z_check_solution(n, n, nrhs, Acpy, lda, b, x, ldb,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free(Acpy); free(b);
}
free(A); free(x);
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
PLASMA_Complex64_t *A, *A2 = NULL;
real_Double_t t;
int *ipiv, *ipiv2 = NULL;
int i;
int m = iparam[TIMING_N];
int n = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = m;
PLASMA_sequence *sequence = NULL;
PLASMA_request request = PLASMA_REQUEST_INITIALIZER;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* Allocate Data */
A = (PLASMA_Complex64_t *)malloc(lda*n*sizeof(PLASMA_Complex64_t));
/* Check if unable to allocate memory */
if ( (! A) ) {
printf("Out of Memory \n ");
return -1;
}
/* Initialiaze Data */
LAPACKE_zlarnv_work(1, ISEED, lda*n, A);
/* Allocate Workspace */
ipiv = (int *)malloc( n*sizeof(int) );
/* Save A in lapack layout for check */
if ( check ) {
A2 = (PLASMA_Complex64_t *)malloc(lda*n*sizeof(PLASMA_Complex64_t));
ipiv2 = (int *)malloc( n*sizeof(int) );
LAPACKE_zlacpy_work(LAPACK_COL_MAJOR,' ', m, n, A, lda, A2, lda);
LAPACKE_zgetrf_work(LAPACK_COL_MAJOR, m, n, A2, lda, ipiv2 );
}
plasma = plasma_context_self();
PLASMA_Sequence_Create(&sequence);
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
QUARK_Task_Flag_Set(&task_flags, TASK_THREAD_COUNT, iparam[TIMING_THRDNBR] );
plasma_dynamic_spawn();
CORE_zgetrf_reclap_init();
t = -cWtime();
QUARK_CORE_zgetrf_reclap(plasma->quark, &task_flags,
m, n, n,
A, lda, ipiv,
sequence, &request,
0, 0,
iparam[TIMING_THRDNBR]);
PLASMA_Sequence_Wait(sequence);
t += cWtime();
*t_ = t;
PLASMA_Sequence_Destroy(sequence);
/* Check the solution */
if ( check )
{
double *work = (double *)malloc(max(m,n)*sizeof(double));
/* Check ipiv */
for(i=0; i<n; i++)
{
if( ipiv[i] != ipiv2[i] ) {
fprintf(stderr, "\nPLASMA (ipiv[%d] = %d, A[%d] = %e) / LAPACK (ipiv[%d] = %d, A[%d] = [%e])\n",
i, ipiv[i], i, creal(A[ i * lda + i ]),
i, ipiv2[i], i, creal(A2[ i * lda + i ]));
break;
}
}
dparam[TIMING_ANORM] = LAPACKE_zlange_work(LAPACK_COL_MAJOR, lapack_const(PlasmaMaxNorm),
m, n, A, lda, work);
dparam[TIMING_XNORM] = LAPACKE_zlange_work(LAPACK_COL_MAJOR, lapack_const(PlasmaMaxNorm),
m, n, A2, lda, work);
dparam[TIMING_BNORM] = 0.0;
CORE_zaxpy( m, n, -1.0, A, lda, A2, lda);
dparam[TIMING_RES] = LAPACKE_zlange_work(LAPACK_COL_MAJOR, lapack_const(PlasmaMaxNorm),
m, n, A2, lda, work);
free( A2 );
free( ipiv2 );
free( work );
}
free( A );
free( ipiv );
PLASMA_Finalize();
return 0;
}
int main( int argc, char *argv[] )
{
int
i, j,
size,
n_threads,
n_repeats,
n_trials,
nb_alg,
increment,
begin;
FLA_Datatype
datatype = FLA_DOUBLE;
FLA_Obj
A;
double
b_norm_value = 0.0,
dtime,
*dtimes,
*flops,
*T;
char
output_file_m[100];
FILE
*fpp;
fprintf( stdout, "%c Enter number of repeats: ", '%' );
scanf( "%d", &n_repeats );
fprintf( stdout, "%c %d\n", '%', n_repeats );
fprintf( stdout, "%c Enter blocksize: ", '%' );
scanf( "%d", &nb_alg );
fprintf( stdout, "%c %d\n", '%', nb_alg );
fprintf( stdout, "%c Enter problem size parameters: first, inc, num: ", '%' );
scanf( "%d%d%d", &begin, &increment, &n_trials );
fprintf( stdout, "%c %d %d %d\n", '%', begin, increment, n_trials );
fprintf( stdout, "%c Enter number of threads: ", '%' );
scanf( "%d", &n_threads );
fprintf( stdout, "%c %d\n\n", '%', n_threads );
sprintf( output_file_m, "%s/%s_output.m", OUTPUT_PATH, OUTPUT_FILE );
fpp = fopen( output_file_m, "a" );
fprintf( fpp, "%%\n" );
fprintf( fpp, "%% | Matrix Size | PLASMA |\n" );
fprintf( fpp, "%% | n x n | GFlops |\n" );
fprintf( fpp, "%% -----------------------------\n" );
FLA_Init();
PLASMA_Init( n_threads );
PLASMA_Disable( PLASMA_AUTOTUNING );
PLASMA_Set( PLASMA_TILE_SIZE, nb_alg );
PLASMA_Set( PLASMA_INNER_BLOCK_SIZE, nb_alg / 4 );
dtimes = ( double * ) FLA_malloc( n_repeats * sizeof( double ) );
flops = ( double * ) FLA_malloc( n_trials * sizeof( double ) );
fprintf( fpp, "%s = [\n", OUTPUT_FILE );
for ( i = 0; i < n_trials; i++ )
{
size = begin + i * increment;
FLA_Obj_create( datatype, size, size, 0, 0, &A );
for ( j = 0; j < n_repeats; j++ )
{
FLA_Random_matrix( A );
PLASMA_Alloc_Workspace_dgeqrf( size, size, &T );
dtime = FLA_Clock();
PLASMA_dgeqrf( size, size, FLA_Obj_buffer_at_view( A ), size, T );
dtime = FLA_Clock() - dtime;
dtimes[j] = dtime;
free( T );
}
dtime = dtimes[0];
for ( j = 1; j < n_repeats; j++ )
dtime = min( dtime, dtimes[j] );
flops[i] = 4.0 / 3.0 * size * size * size / dtime / 1e9;
fprintf( fpp, " %d %6.3f\n", size, flops[i] );
printf( "Time: %e | GFlops: %6.3f\n",
dtime, flops[i] );
printf( "Matrix size: %d x %d | nb_alg: %d\n",
size, size, nb_alg );
printf( "Norm of difference: %le\n\n", b_norm_value );
FLA_Obj_free( &A );
}
fprintf( fpp, "];\n" );
fflush( fpp );
fclose( fpp );
FLA_free( dtimes );
FLA_free( flops );
PLASMA_Finalize();
FLA_Finalize();
return 0;
}
static int
RunTest(int *iparam, _PREC *dparam, real_Double_t *t_)
{
PLASMA_Complex32_t *A, *Acpy = NULL;
real_Double_t t;
int n = iparam[TIMING_N];
int nb = iparam[TIMING_NB];
int check = iparam[TIMING_CHECK];
n = ((n % nb) == 0) ? (n / nb) * nb : ((n / nb) + 1) * nb ;
dparam[TIMING_ANORM] = (_PREC)n;
dparam[TIMING_BNORM] = (_PREC)_FADDS;
/* Allocate Data */
A = (PLASMA_Complex32_t *)malloc(n*n*sizeof(PLASMA_Complex32_t));
/* Check if unable to allocate memory */
if ( (!A) ) {
printf("Out of Memory \n ");
exit(0);
}
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
/* } */
/* Initialiaze Data */
LAPACKE_clarnv_work(1, ISEED, n*n, A);
/* Save A and b */
if (check) {
Acpy = (PLASMA_Complex32_t *)malloc(n*n*sizeof(PLASMA_Complex32_t));
LAPACKE_clacpy_work(LAPACK_COL_MAJOR, lapack_const(PlasmaUpperLower), n, n, A, n, Acpy, n);
}
t = -cWtime();
PLASMA_cgecfi( n, n, A, PlasmaCM, n, 1, PlasmaCCRB, nb, nb);
t += cWtime();
*t_ = t;
/* Check the solution */
if (check)
{
dparam[TIMING_RES] = (_PREC)c_check_conversion(n, n, n, 1, nb, nb, Acpy, A, map_CM, map_CCRB);
free(Acpy);
}
free( A );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
float *AT, *BT, *CT;
float *A = NULL, *B = NULL, *C1 = NULL, *C2 = NULL;
float alpha, beta;
PLASMA_desc *descA, *descB, *descC;
real_Double_t t;
int nb, nb2, nt;
int n = iparam[TIMING_N];
int check = iparam[TIMING_CHECK];
int lda = n;
/* Allocate Data */
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*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);
AT = (float *)malloc(nt*nt*nb2*sizeof(float));
BT = (float *)malloc(nt*nt*nb2*sizeof(float));
CT = (float *)malloc(nt*nt*nb2*sizeof(float));
/* Check if unable to allocate memory */
if ( (!AT) || (!BT) || (!CT) ) {
printf("Out of Memory \n ");
exit(0);
}
#if defined(PLASMA_CUDA)
cudaHostRegister(AT, nt*nt*nb2*sizeof(float), cudaHostRegisterPortable);
cudaHostRegister(BT, nt*nt*nb2*sizeof(float), cudaHostRegisterPortable);
cudaHostRegister(CT, nt*nt*nb2*sizeof(float), cudaHostRegisterPortable);
#endif
/* Initialiaze Data */
LAPACKE_slarnv_work(1, ISEED, 1, &alpha);
LAPACKE_slarnv_work(1, ISEED, 1, &beta);
LAPACKE_slarnv_work(1, ISEED, nt*nt*nb2, AT);
LAPACKE_slarnv_work(1, ISEED, nt*nt*nb2, BT);
LAPACKE_slarnv_work(1, ISEED, nt*nt*nb2, CT);
/* Initialize AT and bT for Symmetric Positif Matrix */
PLASMA_Desc_Create(&descA, AT, PlasmaRealFloat, nb, nb, nb*nb, n, n, 0, 0, n, n);
PLASMA_Desc_Create(&descB, BT, PlasmaRealFloat, nb, nb, nb*nb, n, n, 0, 0, n, n);
PLASMA_Desc_Create(&descC, CT, PlasmaRealFloat, nb, nb, nb*nb, n, n, 0, 0, n, n);
if (check)
{
C2 = (float *)malloc(n*lda*sizeof(float));
PLASMA_Tile_to_Lapack(descC, (void*)C2, n);
}
#if defined(PLASMA_CUDA)
core_cublas_init();
#endif
t = -cWtime();
PLASMA_sgemm_Tile( PlasmaNoTrans, PlasmaNoTrans, alpha, descA, descB, beta, descC );
t += cWtime();
*t_ = t;
/* Check the solution */
if (check)
{
A = (float *)malloc(n*lda*sizeof(float));
PLASMA_Tile_to_Lapack(descA, (void*)A, n);
free(AT);
B = (float *)malloc(n*lda*sizeof(float));
PLASMA_Tile_to_Lapack(descB, (void*)B, n);
free(BT);
C1 = (float *)malloc(n*lda*sizeof(float));
PLASMA_Tile_to_Lapack(descC, (void*)C1, n);
free(CT);
dparam[TIMING_RES] = s_check_gemm( PlasmaNoTrans, PlasmaNoTrans, n, n, n,
alpha, A, lda, B, lda, beta, C1, C2, lda,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free(C2);
}
else {
free( AT );
free( BT );
free( CT );
}
PLASMA_Desc_Destroy(&descA);
PLASMA_Desc_Destroy(&descB);
PLASMA_Desc_Destroy(&descC);
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
double *A, *b, *x;
double *Acpy = NULL;
double *bcpy = NULL;
real_Double_t t;
int *piv;
int n = iparam[TIMING_N];
int nrhs = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = n;
int ldb = n;
int iter = 0;
/* Allocate Data */
A = (double *)malloc(lda*n* sizeof(double));
b = (double *)malloc(ldb*nrhs*sizeof(double));
x = (double *)malloc(ldb*nrhs*sizeof(double));
piv = (int *)malloc( n*sizeof(int));
/* Check if unable to allocate memory */
if ( (!A) || (!b) || (!x) || (!piv) ) {
printf("Out of Memory \n ");
return -1;
}
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } */
/* Initialiaze Data */
LAPACKE_dlarnv_work(1, ISEED, lda*n, A);
LAPACKE_dlarnv_work(1, ISEED, ldb*nrhs, b);
/* Save A and b */
if (check) {
Acpy = (double *)malloc(lda*n* sizeof(double));
bcpy = (double *)malloc(ldb*nrhs*sizeof(double));
LAPACKE_dlacpy_work(LAPACK_COL_MAJOR,' ', n, n, A, lda, Acpy, lda);
LAPACKE_dlacpy_work(LAPACK_COL_MAJOR,' ', n, nrhs, b, ldb, bcpy, ldb);
}
t = -cWtime();
PLASMA_dsgesv( n, nrhs, A, lda, piv, b, ldb, x, ldb, &iter );
t += cWtime();
*t_ = t;
/* Check the solution */
if (check)
{
dparam[TIMING_RES] = d_check_solution(n, n, nrhs, Acpy, lda, bcpy, x, ldb,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free(Acpy); free(bcpy);
}
free( piv );
free( x );
free( b );
free( A );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, _PREC *dparam, real_Double_t *t_)
{
PLASMA_Complex32_t *A = NULL, *AT;
PLASMA_desc *descA;
real_Double_t t;
int n = iparam[TIMING_N];
int nb = iparam[TIMING_NB];
int check = iparam[TIMING_CHECK];
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
/* } */
n = ((n % nb) == 0) ? (n / nb) * nb : ((n / nb) + 1) * nb ;
dparam[TIMING_ANORM] = (_PREC)n;
/* Allocate Data */
AT = (PLASMA_Complex32_t *)malloc(n*n*sizeof(PLASMA_Complex32_t));
/* Check if unable to allocate memory */
if ( (!AT) ) {
printf("Out of Memory \n ");
exit(0);
}
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaComplexFloat, nb, nb, nb*nb, n, n, 0, 0, n, n);
LAPACKE_clarnv_work(1, ISEED, n*n, AT);
/* Save A and b */
if (check) {
A = (PLASMA_Complex32_t *)malloc(n*n*sizeof(PLASMA_Complex32_t));
LAPACKE_clacpy_work(LAPACK_COL_MAJOR, lapack_const(PlasmaUpperLower), n, n, AT, n, A, n);
}
t = -cWtime();
PLASMA_Lapack_to_Tile( (void *)A, n, descA);
t += cWtime();
*t_ = t;
/* Check the solution */
if (check)
{
dparam[TIMING_RES] = (_PREC)c_check_conversion(n, n, n, 1, nb, nb, A, AT, map_CM, map_CCRB);
free(A);
}
PLASMA_Desc_Destroy(&descA);
free( AT );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
PLASMA_Complex64_t *AT;
PLASMA_desc *descA, *descT;
real_Double_t t;
int nb;
int M = iparam[TIMING_N];
int N = iparam[TIMING_M];
//int N = M/nrhs; //RUN WITH NRHS = 10 or 20 (ALSO USED IN TIMING.C)
int lda = M;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
nb = iparam[TIMING_NB];
/* Householder mode */
//PLASMA_Set(PLASMA_HOUSEHOLDER_MODE, PLASMA_FLAT_HOUSEHOLDER);
PLASMA_Set(PLASMA_HOUSEHOLDER_MODE, PLASMA_TREE_HOUSEHOLDER);
PLASMA_Set(PLASMA_HOUSEHOLDER_SIZE, 4);
/* Allocate Data */
AT = (PLASMA_Complex64_t *)malloc(lda*N*sizeof(PLASMA_Complex64_t));
/* Check if unable to allocate memory */
if ( !AT ){
printf("Out of Memory \n ");
exit(0);
}
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaComplexDouble, nb, nb, nb*nb, M, N, 0, 0, M, N);
LAPACKE_zlarnv_work(1, ISEED, lda*N, AT);
/* Allocate Workspace */
PLASMA_Alloc_Workspace_zgels_Tile(M, N, &descT);
t = -cWtime();
PLASMA_zgeqrf_Tile( descA, descT );
t += cWtime();
*t_ = t;
/* Allocate Workspace */
PLASMA_Dealloc_Handle_Tile(&descT);
PLASMA_Desc_Destroy(&descA);
free( AT );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
PLASMA_Complex32_t *A, *Acpy = NULL, *L, *b, *x;
real_Double_t t;
int *piv;
int n = iparam[TIMING_N];
int nrhs = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = n;
int ldb = n;
/* Allocate Data */
A = (PLASMA_Complex32_t *)malloc(lda*n*sizeof(PLASMA_Complex32_t));
/* Check if unable to allocate memory */
if ( !A ){
printf("Out of Memory \n ");
exit(0);
}
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
/* Initialiaze Data */
LAPACKE_clarnv_work(1, ISEED, n*lda, A);
/* Allocate Workspace */
PLASMA_Alloc_Workspace_cgesv_incpiv(n, &L, &piv);
/* Save AT in lapack layout for check */
if ( check ) {
Acpy = (PLASMA_Complex32_t *)malloc(lda*n*sizeof(PLASMA_Complex32_t));
LAPACKE_clacpy_work(LAPACK_COL_MAJOR,' ', n, n, A, lda, Acpy, lda);
}
t = -cWtime();
PLASMA_cgetrf_incpiv( n, n, A, lda, L, piv );
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
b = (PLASMA_Complex32_t *)malloc(ldb*nrhs *sizeof(PLASMA_Complex32_t));
x = (PLASMA_Complex32_t *)malloc(ldb*nrhs *sizeof(PLASMA_Complex32_t));
LAPACKE_clarnv_work(1, ISEED, ldb*nrhs, x);
LAPACKE_clacpy_work(LAPACK_COL_MAJOR,' ', n, nrhs, x, ldb, b, ldb);
PLASMA_cgetrs_incpiv( PlasmaNoTrans, n, nrhs, A, lda, L, piv, x, ldb );
dparam[TIMING_RES] = c_check_solution(n, n, nrhs, Acpy, lda, b, x, ldb,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free( Acpy ); free( b ); free( x );
}
free( A );
free( L );
free( piv );
PLASMA_Finalize();
return 0;
}
int testing_cgels(int argc, char **argv)
{
int mode = 0;
if ( argc < 1 ){
goto usage;
} else {
mode = atoi(argv[0]);
}
/* Check for number of arguments*/
if ( ((mode == 0) && (argc != 6)) ||
((mode != 0) && (argc != 7)) ){
usage:
USAGE("GELS", "MODE M N LDA NRHS LDB [RH]",
" - MODE : 0: flat, 1: tree (RH needed)\n"
" - M : number of rows of the matrix A\n"
" - N : number of columns of the matrix A\n"
" - LDA : leading dimension of the matrix A\n"
" - NRHS : number of RHS\n"
" - LDB : leading dimension of the matrix B\n"
" - RH : Size of each subdomains\n");
return -1;
}
int M = atoi(argv[1]);
int N = atoi(argv[2]);
int LDA = atoi(argv[3]);
int NRHS = atoi(argv[4]);
int LDB = atoi(argv[5]);
int rh;
int K = min(M, N);
float eps;
int info_ortho, info_solution, info_factorization;
int i,j;
int LDAxN = LDA*N;
int LDBxNRHS = LDB*NRHS;
PLASMA_Complex32_t *A1 = (PLASMA_Complex32_t *)malloc(LDA*N*sizeof(PLASMA_Complex32_t));
PLASMA_Complex32_t *A2 = (PLASMA_Complex32_t *)malloc(LDA*N*sizeof(PLASMA_Complex32_t));
PLASMA_Complex32_t *B1 = (PLASMA_Complex32_t *)malloc(LDB*NRHS*sizeof(PLASMA_Complex32_t));
PLASMA_Complex32_t *B2 = (PLASMA_Complex32_t *)malloc(LDB*NRHS*sizeof(PLASMA_Complex32_t));
PLASMA_Complex32_t *Q = (PLASMA_Complex32_t *)malloc(LDA*N*sizeof(PLASMA_Complex32_t));
PLASMA_Complex32_t *T;
/* Check if unable to allocate memory */
if ((!A1)||(!A2)||(!B1)||(!B2)||(!Q)){
printf("Out of Memory \n ");
return -2;
}
if ( mode ) {
rh = atoi(argv[6]);
PLASMA_Set(PLASMA_HOUSEHOLDER_MODE, PLASMA_TREE_HOUSEHOLDER);
PLASMA_Set(PLASMA_HOUSEHOLDER_SIZE, rh);
}
PLASMA_Alloc_Workspace_cgels(M, N, &T);
eps = BLAS_sfpinfo( blas_eps );
/*----------------------------------------------------------
* TESTING CGELS
*/
/* Initialize A1 and A2 */
LAPACKE_clarnv_work(IONE, ISEED, LDAxN, A1);
for (i = 0; i < M; i++)
for (j = 0; j < N; j++)
A2[LDA*j+i] = A1[LDA*j+i] ;
/* Initialize B1 and B2 */
LAPACKE_clarnv_work(IONE, ISEED, LDBxNRHS, B1);
for (i = 0; i < M; i++)
for (j = 0; j < NRHS; j++)
B2[LDB*j+i] = B1[LDB*j+i] ;
memset((void*)Q, 0, LDA*N*sizeof(PLASMA_Complex32_t));
for (i = 0; i < K; i++)
Q[LDA*i+i] = 1.0;
/* PLASMA CGELS */
PLASMA_cgels(PlasmaNoTrans, M, N, NRHS, A2, LDA, T, B2, LDB);
/* PLASMA CGELS */
if (M >= N)
/* Building the economy-size Q */
PLASMA_cungqr(M, N, K, A2, LDA, T, Q, LDA);
else
/* Building the economy-size Q */
PLASMA_cunglq(M, N, K, A2, LDA, T, Q, LDA);
printf("\n");
printf("------ TESTS FOR PLASMA CGELS ROUTINE ------- \n");
printf(" Size of the Matrix %d by %d\n", M, N);
printf("\n");
printf(" The matrix A is randomly generated for each test.\n");
printf("============\n");
printf(" The relative machine precision (eps) is to be %e \n",eps);
printf(" Computational tests pass if scaled residuals are less than 60.\n");
/* Check the orthogonality, factorization and the solution */
info_ortho = check_orthogonality(M, N, LDA, Q, eps);
info_factorization = check_factorization(M, N, A1, A2, LDA, Q, eps);
info_solution = check_solution(M, N, NRHS, A1, LDA, B1, B2, LDB, eps);
if ((info_solution == 0)&(info_factorization == 0)&(info_ortho == 0)) {
printf("***************************************************\n");
printf(" ---- TESTING CGELS ...................... PASSED !\n");
printf("***************************************************\n");
}
else {
printf("************************************************\n");
printf(" - TESTING CGELS ... FAILED !\n");
printf("************************************************\n");
}
/*-------------------------------------------------------------
* TESTING CGEQRF + CGEQRS or CGELQF + CGELQS
*/
/* Initialize A1 and A2 */
LAPACKE_clarnv_work(IONE, ISEED, LDAxN, A1);
for (i = 0; i < M; i++)
for (j = 0; j < N; j++)
A2[LDA*j+i] = A1[LDA*j+i];
/* Initialize B1 and B2 */
LAPACKE_clarnv_work(IONE, ISEED, LDBxNRHS, B1);
for (i = 0; i < M; i++)
for (j = 0; j < NRHS; j++)
B2[LDB*j+i] = B1[LDB*j+i];
memset((void*)Q, 0, LDA*N*sizeof(PLASMA_Complex32_t));
for (i = 0; i < K; i++)
Q[LDA*i+i] = 1.0;
if (M >= N) {
printf("\n");
printf("------ TESTS FOR PLASMA CGEQRF + CGEQRS ROUTINE ------- \n");
printf(" Size of the Matrix %d by %d\n", M, N);
printf("\n");
printf(" The matrix A is randomly generated for each test.\n");
printf("============\n");
printf(" The relative machine precision (eps) is to be %e \n", eps);
printf(" Computational tests pass if scaled residuals are less than 60.\n");
/* Plasma routines */
PLASMA_cgeqrf(M, N, A2, LDA, T);
PLASMA_cungqr(M, N, K, A2, LDA, T, Q, LDA);
PLASMA_cgeqrs(M, N, NRHS, A2, LDA, T, B2, LDB);
/* Check the orthogonality, factorization and the solution */
info_ortho = check_orthogonality(M, N, LDA, Q, eps);
info_factorization = check_factorization(M, N, A1, A2, LDA, Q, eps);
info_solution = check_solution(M, N, NRHS, A1, LDA, B1, B2, LDB, eps);
if ((info_solution == 0)&(info_factorization == 0)&(info_ortho == 0)) {
printf("***************************************************\n");
printf(" ---- TESTING CGEQRF + CGEQRS ............ PASSED !\n");
printf("***************************************************\n");
}
else{
printf("***************************************************\n");
printf(" - TESTING CGEQRF + CGEQRS ... FAILED !\n");
printf("***************************************************\n");
}
}
else {
printf("\n");
printf("------ TESTS FOR PLASMA CGELQF + CGELQS ROUTINE ------- \n");
printf(" Size of the Matrix %d by %d\n", M, N);
printf("\n");
printf(" The matrix A is randomly generated for each test.\n");
printf("============\n");
printf(" The relative machine precision (eps) is to be %e \n", eps);
printf(" Computational tests pass if scaled residuals are less than 60.\n");
/* Plasma routines */
PLASMA_cgelqf(M, N, A2, LDA, T);
PLASMA_cunglq(M, N, K, A2, LDA, T, Q, LDA);
PLASMA_cgelqs(M, N, NRHS, A2, LDA, T, B2, LDB);
/* Check the orthogonality, factorization and the solution */
info_ortho = check_orthogonality(M, N, LDA, Q, eps);
info_factorization = check_factorization(M, N, A1, A2, LDA, Q, eps);
info_solution = check_solution(M, N, NRHS, A1, LDA, B1, B2, LDB, eps);
if ( (info_solution == 0) & (info_factorization == 0) & (info_ortho == 0) ) {
printf("***************************************************\n");
printf(" ---- TESTING CGELQF + CGELQS ............ PASSED !\n");
printf("***************************************************\n");
}
else {
printf("***************************************************\n");
printf(" - TESTING CGELQF + CGELQS ... FAILED !\n");
printf("***************************************************\n");
}
}
/*----------------------------------------------------------
* TESTING CGEQRF + ZORMQR + CTRSM
*/
/* Initialize A1 and A2 */
LAPACKE_clarnv_work(IONE, ISEED, LDAxN, A1);
for (i = 0; i < M; i++)
for (j = 0; j < N; j++)
A2[LDA*j+i] = A1[LDA*j+i];
/* Initialize B1 and B2 */
memset(B2, 0, LDB*NRHS*sizeof(PLASMA_Complex32_t));
LAPACKE_clarnv_work(IONE, ISEED, LDBxNRHS, B1);
for (i = 0; i < M; i++)
for (j = 0; j < NRHS; j++)
B2[LDB*j+i] = B1[LDB*j+i];
/* PLASMA CGEQRF+ CUNMQR + CTRSM */
memset((void*)Q, 0, LDA*N*sizeof(PLASMA_Complex32_t));
for (i = 0; i < K; i++)
Q[LDA*i+i] = 1.0;
if (M >= N) {
printf("\n");
printf("------ TESTS FOR PLASMA CGEQRF + CUNMQR + CTRSM ROUTINE ------- \n");
printf(" Size of the Matrix %d by %d\n", M, N);
printf("\n");
printf(" The matrix A is randomly generated for each test.\n");
printf("============\n");
printf(" The relative machine precision (eps) is to be %e \n",eps);
printf(" Computational tests pass if scaled residuals are less than 60.\n");
PLASMA_cgeqrf(M, N, A2, LDA, T);
PLASMA_cungqr(M, N, K, A2, LDA, T, Q, LDA);
PLASMA_cunmqr(PlasmaLeft, PlasmaConjTrans, M, NRHS, N, A2, LDA, T, B2, LDB);
PLASMA_ctrsm(PlasmaLeft, PlasmaUpper, PlasmaNoTrans, PlasmaNonUnit, N, NRHS, 1.0, A2, LDA, B2, LDB);
}
else {
printf("\n");
printf("------ TESTS FOR PLASMA CGELQF + CUNMLQ + CTRSM ROUTINE ------- \n");
printf(" Size of the Matrix %d by %d\n", M, N);
printf("\n");
printf(" The matrix A is randomly generated for each test.\n");
printf("============\n");
printf(" The relative machine precision (eps) is to be %e \n",eps);
printf(" Computational tests pass if scaled residuals are less than 60.\n");
PLASMA_cgelqf(M, N, A2, LDA, T);
PLASMA_ctrsm(PlasmaLeft, PlasmaLower, PlasmaNoTrans, PlasmaNonUnit, M, NRHS, 1.0, A2, LDA, B2, LDB);
PLASMA_cunglq(M, N, K, A2, LDA, T, Q, LDA);
PLASMA_cunmlq(PlasmaLeft, PlasmaConjTrans, N, NRHS, M, A2, LDA, T, B2, LDB);
}
/* Check the orthogonality, factorization and the solution */
info_ortho = check_orthogonality(M, N, LDA, Q, eps);
info_factorization = check_factorization(M, N, A1, A2, LDA, Q, eps);
info_solution = check_solution(M, N, NRHS, A1, LDA, B1, B2, LDB, eps);
if ( (info_solution == 0) & (info_factorization == 0) & (info_ortho == 0) ) {
if (M >= N) {
printf("***************************************************\n");
printf(" ---- TESTING CGEQRF + CUNMQR + CTRSM .... PASSED !\n");
printf("***************************************************\n");
}
else {
printf("***************************************************\n");
printf(" ---- TESTING CGELQF + CTRSM + CUNMLQ .... PASSED !\n");
printf("***************************************************\n");
}
}
else {
if (M >= N) {
printf("***************************************************\n");
printf(" - TESTING CGEQRF + CUNMQR + CTRSM ... FAILED !\n");
printf("***************************************************\n");
}
else {
printf("***************************************************\n");
printf(" - TESTING CGELQF + CTRSM + CUNMLQ ... FAILED !\n");
printf("***************************************************\n");
}
}
free(A1); free(A2); free(B1); free(B2); free(Q); free(T);
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
float *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] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*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 */
AT = (float *)malloc(nt*nt*nb2*sizeof(float));
/* Check if unable to allocate memory */
if ( !AT ){
printf("Out of Memory \n ");
exit(0);
}
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaRealFloat, nb, nb, nb*nb, n, n, 0, 0, n, n);
PLASMA_splgsy_Tile((float)n, descA, 51 );
/* Save AT in lapack layout for check */
if ( check ) {
A = (float *)malloc(lda*n *sizeof(float));
PLASMA_Tile_to_Lapack(descA, (void*)A, n);
}
/* PLASMA SPOSV */
t = -cWtime();
PLASMA_spotrf_Tile(PlasmaUpper, descA);
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
b = (float *)malloc(ldb*nrhs *sizeof(float));
bT = (float *)malloc(nt*nb2 *sizeof(float));
x = (float *)malloc(ldb*nrhs *sizeof(float));
LAPACKE_slarnv_work(1, ISEED, nt*nb2, bT);
PLASMA_Desc_Create(&descB, bT, PlasmaRealFloat, nb, nb, nb*nb, n, nrhs, 0, 0, n, nrhs);
PLASMA_Tile_to_Lapack(descB, (void*)b, n);
PLASMA_spotrs_Tile( PlasmaUpper, descA, descB );
PLASMA_Tile_to_Lapack(descB, (void*)x, n);
dparam[TIMING_RES] = s_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);
free(AT);
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
double *A = NULL, *AT, *b = NULL, *bT, *x;
PLASMA_desc *descA, *descB, *descT;
real_Double_t t;
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] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_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] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
nb = iparam[TIMING_NB];
nb2 = nb * nb;
nt = n / nb + ((n % nb == 0) ? 0 : 1);
/* Allocate Data */
AT = (double *)malloc(nt*nt*nb2*sizeof(double));
/* Check if unable to allocate memory */
if ( !AT ){
printf("Out of Memory \n ");
exit(0);
}
#if defined(PLASMA_CUDA)
cudaHostRegister((void*)AT, nt*nt*nb2*sizeof(double), cudaHostRegisterPortable);
#endif
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaRealDouble, nb, nb, nb*nb, n, n, 0, 0, n, n);
LAPACKE_dlarnv_work(1, ISEED, nt*nt*nb2, AT);
/* Allocate Workspace */
PLASMA_Alloc_Workspace_dgels_Tile(n, n, &descT);
#if defined(PLASMA_CUDA)
cudaHostRegister((void*)descT->mat, descT->lm*descT->ln*sizeof(double), cudaHostRegisterPortable);
#endif
/* Save AT in lapack layout for check */
if ( check ) {
A = (double *)malloc(lda*n *sizeof(double));
PLASMA_Tile_to_Lapack(descA, (void*)A, n);
}
t = -cWtime();
PLASMA_dgeqrf_Tile( descA, descT );
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_dgeqrs_Tile( descA, descT, 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 );
}
/* Allocate Workspace */
PLASMA_Dealloc_Handle_Tile(&descT);
PLASMA_Desc_Destroy(&descA);
free( AT );
PLASMA_Finalize();
#if defined(PLASMA_CUDA)
#endif
return 0;
}
int main (int argc, char **argv)
{
int ncores, sched;
int info;
char func[32];
/* Check for number of arguments*/
if ( argc < 4) {
printf(" Proper Usage is : ./stesting ncores sched FUNC ...\n"
" - ncores : number of cores \n"
" - sched : 0 for static, 1 for dynamic\n"
" - FUNC : name of function to test\n");
exit(1);
}
sscanf( argv[1], "%d", &ncores );
sscanf( argv[2], "%d", &sched );
sscanf( argv[3], "%s", func );
PLASMA_Init(ncores);
if ( sched == 0 )
PLASMA_Set( PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
else
PLASMA_Set( PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
argc -= 4;
argv += 4;
info = 0;
/*
* Norms
*/
if ( strcmp(func, "LANGE") == 0 ) {
info = testing_slange( argc, argv );
/*
* Blas Level 3
*/
} else if ( strcmp(func, "GEMM") == 0 ) {
info = testing_sgemm( argc, argv );
#ifdef COMPLEX
} else if ( strcmp(func, "HEMM") == 0 ) {
info = testing_ssymm( argc, argv );
} else if ( strcmp(func, "HERK") == 0 ) {
info = testing_ssyrk( argc, argv );
} else if ( strcmp(func, "HER2K") == 0 ) {
info = testing_ssyr2k( argc, argv );
#endif
} else if ( strcmp(func, "SYMM") == 0 ) {
info = testing_ssymm( argc, argv );
} else if ( strcmp(func, "SYRK") == 0 ) {
info = testing_ssyrk( argc, argv );
} else if ( strcmp(func, "SYR2K") == 0 ) {
info = testing_ssyr2k( argc, argv );
} else if ( strcmp(func, "TRMM") == 0 ) {
info = testing_strmm( argc, argv );
} else if ( strcmp(func, "TRSM") == 0 ) {
info = testing_strsm( argc, argv );
/*
* Linear system
*/
} else if ( strcmp(func, "POSV") == 0 ) {
info = testing_sposv( argc, argv );
} else if ( strcmp(func, "GELS") == 0 ) {
info = testing_sgels( argc, argv );
} else if ( strcmp(func, "GESV") == 0 ) {
info = testing_sgesv( argc, argv );
/*
* Eigenvalue Problems
*/
} else if ( strcmp(func, "HEEV") == 0 ) {
info = testing_ssyev( argc, argv );
} else if ( strcmp(func, "HEGV") == 0 ) {
info = testing_ssygv( argc, argv );
} else if ( strcmp(func, "HEGST") == 0 ) {
info = testing_ssygst( argc, argv );
/*
* Singular Value Decomposition
*/
} else if ( strcmp(func, "GESVD") == 0 ) {
info = testing_sgesvd( argc, argv );
#ifdef DOUBLE
/*
* Mixed precision
*/
} else if ( strcmp(func, "SPOSV") == 0 ) {
info = testing_dsposv( argc, argv );
} else if ( strcmp(func, "SGESV") == 0 ) {
info = testing_dsgesv( argc, argv );
} else if ( strcmp(func, "SUNGESV") == 0 ) {
info = testing_dsungesv( argc, argv );
#endif
/* Layout Transformation */
} else if ( strcmp(func, "GECFI") == 0 ) {
info = testing_sgecfi( argc, argv );
} else if ( strcmp(func, "GETMI") == 0 ) {
info = testing_sgetmi( argc, argv );
} else {
fprintf(stderr, "Function unknown\n");
}
if ( info == -1 ) {
printf( "TESTING %s FAILED : incorrect number of arguments\n", func);
} else if ( info == -2 ) {
printf( "TESTING %s FAILED : not enough memory\n", func);
}
PLASMA_Finalize();
return EXIT_SUCCESS;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
PLASMA_Complex64_t *A, *B1, *B2 = NULL;
PLASMA_Complex64_t alpha;
real_Double_t t;
int n = iparam[TIMING_N];
int nrhs = n;
int check = iparam[TIMING_CHECK];
int lda = n;
/* Allocate Data */
A = (PLASMA_Complex64_t *)malloc(lda*n *sizeof(PLASMA_Complex64_t));
B1 = (PLASMA_Complex64_t *)malloc(lda*nrhs*sizeof(PLASMA_Complex64_t));
/* Check if unable to allocate memory */
if ( (!A) || (!B1) ) {
printf("Out of Memory \n ");
exit(0);
}
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
/* } */
/* Initialiaze Data */
lapack_zlarnv(1, ISEED, n *lda, A );
lapack_zlarnv(1, ISEED, nrhs*lda, B1);
lapack_zlarnv(1, ISEED, 1, &alpha);
int i;
for(i=0;i<max(n, nrhs);i++)
A[lda*i+i] = A[lda*i+i] + 2.0;
if (check)
{
B2 = (PLASMA_Complex64_t *)malloc(lda*nrhs*sizeof(PLASMA_Complex64_t));
memcpy(B2, B1, lda*nrhs*sizeof(PLASMA_Complex64_t));
}
t = -cWtime();
PLASMA_ztrsm( PlasmaLeft, PlasmaUpper, PlasmaNoTrans, PlasmaUnit,
n, nrhs, alpha, A, lda, B1, lda );
t += cWtime();
*t_ = t;
/* Check the solution */
if (check)
{
dparam[TIMING_RES] = z_check_trsm( PlasmaLeft, PlasmaUpper, PlasmaNoTrans, PlasmaUnit, n, nrhs,
alpha, A, lda, B1, B2, lda,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free(B2);
}
free( A );
free( B1 );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
PLASMA_Complex64_t *A = NULL, *AT, *b, *bT, *x;
PLASMA_desc *descA, *descB, *descL;
real_Double_t t;
int *piv;
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] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
nb = iparam[TIMING_NB];
nb2 = nb * nb;
nt = n / nb + ((n % nb == 0) ? 0 : 1);
/* Allocate Data */
AT = (PLASMA_Complex64_t *)malloc(nt*nt*nb2*sizeof(PLASMA_Complex64_t));
/* Check if unable to allocate memory */
if ( !AT ){
printf("Out of Memory \n ");
exit(0);
}
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaComplexDouble, nb, nb, nb*nb, n, n, 0, 0, n, n);
LAPACKE_zlarnv_work(1, ISEED, nt*nt*nb2, AT);
/* Allocate Workspace */
PLASMA_Alloc_Workspace_zgesv_incpiv_Tile(n, &descL, &piv);
/* Save AT in lapack layout for check */
if ( check ) {
A = (PLASMA_Complex64_t *)malloc(lda*n *sizeof(PLASMA_Complex64_t));
PLASMA_Tile_to_Lapack(descA, (void*)A, n);
}
t = -cWtime();
PLASMA_zgetrf_incpiv_Tile( descA, descL, piv );
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
b = (PLASMA_Complex64_t *)malloc(ldb*nrhs *sizeof(PLASMA_Complex64_t));
bT = (PLASMA_Complex64_t *)malloc(nt*nb2 *sizeof(PLASMA_Complex64_t));
x = (PLASMA_Complex64_t *)malloc(ldb*nrhs *sizeof(PLASMA_Complex64_t));
LAPACKE_zlarnv_work(1, ISEED, n*nrhs, b);
PLASMA_Desc_Create(&descB, bT, PlasmaComplexDouble, nb, nb, nb*nb, n, nrhs, 0, 0, n, nrhs);
PLASMA_Lapack_to_Tile((void*)b, n, descB);
PLASMA_zgetrs_incpiv_Tile( descA, descL, piv, descB );
PLASMA_Tile_to_Lapack(descB, (void*)x, n);
dparam[TIMING_RES] = z_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 );
}
/* Deallocate Workspace */
PLASMA_Dealloc_Handle_Tile(&descL);
PLASMA_Desc_Destroy(&descA);
free( AT );
free( piv );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
float *AT, *bT, *x;
float *A = NULL;
float *b = NULL;
PLASMA_desc *descA, *descB;
real_Double_t t;
int *piv;
int n = iparam[TIMING_N];
int nb = iparam[TIMING_NB];
int nrhs = iparam[TIMING_NRHS];
int check = iparam[TIMING_CHECK];
int lda = n;
int ldb = n;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
/* Allocate Data */
AT = (float *)malloc(lda*n *sizeof(float));
bT = (float *)malloc(ldb*nrhs*sizeof(float));
piv = (int *)malloc( n*sizeof(int));
/* Check if unable to allocate memory */
if ( (!AT) || (!bT) || (!piv) ) {
printf("Out of Memory \n ");
return -1;
}
/* Initialize AT and bT for Symmetric Positif Matrix */
PLASMA_Desc_Create(&descA, AT, PlasmaRealFloat, nb, nb, nb*nb, lda, n, 0, 0, n, n);
PLASMA_Desc_Create(&descB, bT, PlasmaRealFloat, nb, nb, nb*nb, ldb, nrhs, 0, 0, n, nrhs);
LAPACKE_slarnv_work(1, ISEED, lda*n, AT);
LAPACKE_slarnv_work(1, ISEED, ldb*nrhs, bT);
/* Save AT and bT in lapack layout for check */
if ( check ) {
A = (float *)malloc(lda*n *sizeof(float));
b = (float *)malloc(ldb*nrhs*sizeof(float));
PLASMA_sTile_to_Lapack(descA, (void*)A, lda);
PLASMA_sTile_to_Lapack(descB, (void*)b, ldb);
}
t = -cWtime();
PLASMA_sgesv_Tile( descA, piv, descB );
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
x = (float *)malloc(ldb*nrhs *sizeof(float));
PLASMA_sTile_to_Lapack(descB, (void*)x, n);
dparam[TIMING_RES] = s_check_solution(n, n, nrhs, A, lda, b, x, ldb,
&(dparam[TIMING_ANORM]), &(dparam[TIMING_BNORM]),
&(dparam[TIMING_XNORM]));
free(A); free(b); free(x);
}
PLASMA_Desc_Destroy(&descA);
PLASMA_Desc_Destroy(&descB);
free( AT ); free( bT );
free( piv );
PLASMA_Finalize();
return 0;
}
static int
RunTest(int *iparam, double *dparam, real_Double_t *t_)
{
double *AT;
real_Double_t t;
PLASMA_desc *descA;
int nb, nb2, nt;
int n = iparam[TIMING_N];
int check = iparam[TIMING_CHECK];
PLASMA_enum uplo = PlasmaLower;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
/*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 */
AT = (double *)malloc(nt*nt*nb2*sizeof(double));
/* Check if unable to allocate memory */
if ( !AT ){
printf("Out of Memory \n ");
exit(0);
}
/*
* Initialize Data
* It's done in static to avoid having the same sequence than one
* the function we want to trace
*/
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
PLASMA_Desc_Create(&descA, AT, PlasmaRealDouble, nb, nb, nb*nb, n, n, 0, 0, n, n);
PLASMA_dplgsy_Tile( (double)n, descA, 51 );
if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
else
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING );
/* Save AT in lapack layout for check */
if ( check ) {
}
/* PLASMA DPOTRF / DTRTRI / DLAUUM */
/*
* Example of the different way to combine several asynchonous calls
*/
{
#if defined(TRACE_BY_SEQUENCE)
PLASMA_sequence *sequence[3];
PLASMA_request request[3] = { PLASMA_REQUEST_INITIALIZER,
PLASMA_REQUEST_INITIALIZER,
PLASMA_REQUEST_INITIALIZER };
PLASMA_Sequence_Create(&sequence[0]);
PLASMA_Sequence_Create(&sequence[1]);
PLASMA_Sequence_Create(&sequence[2]);
t = -cWtime();
#if defined(POTRI_SYNC)
PLASMA_dpotrf_Tile_Async(uplo, descA, sequence[0], &request[0]);
PLASMA_Sequence_Wait(sequence[0]);
PLASMA_dtrtri_Tile_Async(uplo, PlasmaNonUnit, descA, sequence[1], &request[1]);
PLASMA_Sequence_Wait(sequence[1]);
PLASMA_dlauum_Tile_Async(uplo, descA, sequence[2], &request[2]);
PLASMA_Sequence_Wait(sequence[2]);
#else
PLASMA_dpotrf_Tile_Async(uplo, descA, sequence[0], &request[0]);
PLASMA_dtrtri_Tile_Async(uplo, PlasmaNonUnit, descA, sequence[1], &request[1]);
PLASMA_dlauum_Tile_Async(uplo, descA, sequence[2], &request[2]);
PLASMA_Sequence_Wait(sequence[0]);
PLASMA_Sequence_Wait(sequence[1]);
PLASMA_Sequence_Wait(sequence[2]);
#endif
t += cWtime();
PLASMA_Sequence_Destroy(sequence[0]);
PLASMA_Sequence_Destroy(sequence[1]);
PLASMA_Sequence_Destroy(sequence[2]);
#else
#if defined(POTRI_SYNC)
t = -cWtime();
PLASMA_dpotrf_Tile(uplo, descA);
PLASMA_dtrtri_Tile(uplo, PlasmaNonUnit, descA);
PLASMA_dlauum_Tile(uplo, descA);
t += cWtime();
#else
/* Default: we use Asynchonous call with only one sequence */
PLASMA_sequence *sequence;
PLASMA_request request[2] = { PLASMA_REQUEST_INITIALIZER,
PLASMA_REQUEST_INITIALIZER };
t = -cWtime();
PLASMA_Sequence_Create(&sequence);
PLASMA_dpotrf_Tile_Async(uplo, descA, sequence, &request[0]);
PLASMA_dpotri_Tile_Async(uplo, descA, sequence, &request[1]);
PLASMA_Sequence_Wait(sequence);
t += cWtime();
PLASMA_Sequence_Destroy(sequence);
#endif
#endif
*t_ = t;
}
/* Check the solution */
if ( check )
{
dparam[TIMING_ANORM] = 0.0;
dparam[TIMING_XNORM] = 0.0;
dparam[TIMING_BNORM] = 0.0;
dparam[TIMING_RES] = 0.0;
}
PLASMA_Desc_Destroy(&descA);
PLASMA_Finalize();
free(AT);
return 0;
}
static int
RunTest(int *iparam, float *dparam, real_Double_t *t_)
{
PLASMA_Complex32_t *AT, *BT, *Q = NULL;
float *W;
PLASMA_desc *descA, *descB, *descQ, *descT;
real_Double_t t;
int nb, nb2, nt;
int n = iparam[TIMING_N];
int check = iparam[TIMING_CHECK];
int lda = n;
int itype = 1;
int vec = PlasmaNoVec;
int uplo = PlasmaUpper;
/* Initialize Plasma */
PLASMA_Init( iparam[TIMING_THRDNBR] );
// if ( iparam[TIMING_SCHEDULER] )
PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_DYNAMIC_SCHEDULING );
/* else */
/* PLASMA_Set(PLASMA_SCHEDULING_MODE, PLASMA_STATIC_SCHEDULING ); */
/*if ( !iparam[TIMING_AUTOTUNING] ) {*/
PLASMA_Disable(PLASMA_AUTOTUNING);
PLASMA_Set(PLASMA_TILE_SIZE, iparam[TIMING_NB] );
PLASMA_Set(PLASMA_INNER_BLOCK_SIZE, iparam[TIMING_IB] );
/* } else { */
/* PLASMA_Get(PLASMA_TILE_SIZE, &iparam[TIMING_NB] ); */
/* PLASMA_Get(PLASMA_INNER_BLOCK_SIZE, &iparam[TIMING_IB] ); */
/* } */
nb = iparam[TIMING_NB];
nb2 = nb * nb;
nt = n / nb + ((n % nb == 0) ? 0 : 1);
/* Allocate Data */
AT = (PLASMA_Complex32_t *)malloc(nt*nt*nb2*sizeof(PLASMA_Complex32_t));
BT = (PLASMA_Complex32_t *)malloc(nt*nt*nb2*sizeof(PLASMA_Complex32_t));
W = (float *)malloc(n*sizeof(float));
if (vec == PlasmaVec){
Q = (PLASMA_Complex32_t *)malloc(nt*nt*nb2*sizeof(PLASMA_Complex32_t));
if ( (!Q) ) {
printf("Out of Memory -Q-\n ");
exit(0);
}
}
/* Check if unable to allocate memory */
if ( (!AT) || (!BT) || (!W) ) {
printf("Out of Memory -\n ");
exit(0);
}
/* Initialiaze Data */
PLASMA_Desc_Create(&descA, AT, PlasmaComplexFloat, nb, nb, nb*nb, lda, n, 0, 0, n, n);
PLASMA_cplghe_Tile((float)0.0, descA, 51 );
PLASMA_Desc_Create(&descB, BT, PlasmaComplexFloat, nb, nb, nb*nb, lda, n, 0, 0, n, n);
PLASMA_cplghe_Tile((float)n, descB, 51 );
PLASMA_Desc_Create(&descQ, Q, PlasmaComplexFloat, nb, nb, nb*nb, lda, n, 0, 0, n, n);
/* Save AT and bT in lapack layout for check */
if ( check ) {
}
/* Allocate Workspace */
PLASMA_Alloc_Workspace_chegv(n, n, &descT);
t = -cWtime();
PLASMA_chegv_Tile( itype, vec, uplo, descA, descB, W, descT, descQ );
t += cWtime();
*t_ = t;
/* Check the solution */
if ( check )
{
}
/* DeAllocate Workspace */
PLASMA_Dealloc_Handle_Tile(&descT);
PLASMA_Desc_Destroy(&descA);
PLASMA_Desc_Destroy(&descB);
PLASMA_Desc_Destroy(&descQ);
if (vec == PlasmaVec)
free( Q );
free( AT );
free( W );
PLASMA_Finalize();
return 0;
}