//=============================================================================
int Epetra_SerialSpdDenseSolver::Invert(void)
{
if (!Factored()) Factor(); // Need matrix factored.
POTRI ( SymMatrix_->UPLO(), N_, AF_, LDAF_, &INFO_);
// Copy lower/upper triangle to upper/lower triangle: make full inverse
SymMatrix_->CopyUPLOMat(SymMatrix_->Upper(), AF_, LDAF_, N_);
double DN = N_;
UpdateFlops((DN*DN*DN));
Inverted_ = true;
Factored_ = false;
EPETRA_CHK_ERR(INFO_);
return(0);
}
int INV_CHOL(MATRIX_T L)
{
/* computes the inverse of a symmetric square NxN
* matrix using its Cholesky factorization (in place) */
int info;
int N = (int)L.nrows;
char uplo = 'L';
POTRI(&uplo,&N,L.m,&N,&info);
for(int i=0;i<N;i++){
for(int j=i+1;j<N;j++){
L.m[i+N*j] = L.m[j+N*i];
}}
return info;
}
int main(int argc, char *argv[]){
#ifndef COMPLEX
char *trans[] = {"T", "N"};
#else
char *trans[] = {"C", "N"};
#endif
char *uplo[] = {"U", "L"};
FLOAT alpha[] = {1.0, 0.0};
FLOAT beta [] = {0.0, 0.0};
FLOAT *a, *b;
char *p;
char btest = 'F';
blasint m, i, j, info, uplos=0;
double flops;
int from = 1;
int to = 200;
int step = 1;
struct timeval start, stop;
double time1;
argc--;argv++;
if (argc > 0) { from = atol(*argv); argc--; argv++;}
if (argc > 0) { to = MAX(atol(*argv), from); argc--; argv++;}
if (argc > 0) { step = atol(*argv); argc--; argv++;}
if ((p = getenv("OPENBLAS_UPLO")))
if (*p == 'L') uplos=1;
if ((p = getenv("OPENBLAS_TEST"))) btest=*p;
fprintf(stderr, "From : %3d To : %3d Step = %3d Uplo = %c\n", from, to, step,*uplo[uplos]);
if (( a = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
fprintf(stderr,"Out of Memory!!\n");exit(1);
}
if (( b = (FLOAT *)malloc(sizeof(FLOAT) * to * to * COMPSIZE)) == NULL){
fprintf(stderr,"Out of Memory!!\n");exit(1);
}
for(m = from; m <= to; m += step){
#ifndef COMPLEX
if (uplos & 1) {
for (j = 0; j < m; j++) {
for(i = 0; i < j; i++) a[i + j * m] = 0.;
a[j + j * m] = ((double) rand() / (double) RAND_MAX) + 8.;
for(i = j + 1; i < m; i++) a[i + j * m] = ((double) rand() / (double) RAND_MAX) - 0.5;
}
} else {
for (j = 0; j < m; j++) {
for(i = 0; i < j; i++) a[i + j * m] = ((double) rand() / (double) RAND_MAX) - 0.5;
a[j + j * m] = ((double) rand() / (double) RAND_MAX) + 8.;
for(i = j + 1; i < m; i++) a[i + j * m] = 0.;
}
}
#else
if (uplos & 1) {
for (j = 0; j < m; j++) {
for(i = 0; i < j; i++) {
a[(i + j * m) * 2 + 0] = 0.;
a[(i + j * m) * 2 + 1] = 0.;
}
a[(j + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
a[(j + j * m) * 2 + 1] = 0.;
for(i = j + 1; i < m; i++) {
a[(i + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
a[(i + j * m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
}
}
} else {
for (j = 0; j < m; j++) {
for(i = 0; i < j; i++) {
a[(i + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) - 0.5;
a[(i + j * m) * 2 + 1] = ((double) rand() / (double) RAND_MAX) - 0.5;
}
a[(j + j * m) * 2 + 0] = ((double) rand() / (double) RAND_MAX) + 8.;
a[(j + j * m) * 2 + 1] = 0.;
for(i = j + 1; i < m; i++) {
a[(i + j * m) * 2 + 0] = 0.;
a[(i + j * m) * 2 + 1] = 0.;
}
}
}
#endif
SYRK(uplo[uplos], trans[uplos], &m, &m, alpha, a, &m, beta, b, &m);
gettimeofday( &start, (struct timezone *)0);
POTRF(uplo[uplos], &m, b, &m, &info);
gettimeofday( &stop, (struct timezone *)0);
if (info != 0) {
fprintf(stderr, "Potrf info = %d\n", info);
exit(1);
}
time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
flops = COMPSIZE * COMPSIZE * (1.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 1.0/6.0* (double)m) / time1 * 1.e-6;
if ( btest == 'S' )
{
for(j = 0; j < to; j++){
for(i = 0; i < to * COMPSIZE; i++){
a[i + j * to * COMPSIZE] = ((FLOAT) rand() / (FLOAT) RAND_MAX) - 0.5;
}
}
gettimeofday( &start, (struct timezone *)0);
POTRS(uplo[uplos], &m, &m, b, &m, a, &m, &info);
gettimeofday( &stop, (struct timezone *)0);
if (info != 0) {
fprintf(stderr, "Potrs info = %d\n", info);
exit(1);
}
time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
flops = COMPSIZE * COMPSIZE * (2.0 * (double)m * (double)m *(double)m ) / time1 * 1.e-6;
}
if ( btest == 'I' )
{
gettimeofday( &start, (struct timezone *)0);
POTRI(uplo[uplos], &m, b, &m, &info);
gettimeofday( &stop, (struct timezone *)0);
if (info != 0) {
fprintf(stderr, "Potri info = %d\n", info);
exit(1);
}
time1 = (double)(stop.tv_sec - start.tv_sec) + (double)((stop.tv_usec - start.tv_usec)) * 1.e-6;
flops = COMPSIZE * COMPSIZE * (2.0/3.0 * (double)m * (double)m *(double)m +1.0/2.0* (double)m *(double)m + 5.0/6.0* (double)m) / time1 * 1.e-6;
}
fprintf(stderr, "%8d : %10.2f MFlops : %10.3f Sec : Test=%c\n",m,flops ,time1,btest);
}
return 0;
}
