JNIEXPORT jint JNICALL Java_edu_berkeley_bid_SPBLAS_dcsrmm
(JNIEnv * env, jobject calling_obj, jstring j_transa, jint m, jint n, jint k, jdouble alpha, jstring j_matdescra,
jdoubleArray j_vals, jintArray j_ir, jintArray j_jc, jdoubleArray j_b, jint ldb, jdouble beta, jdoubleArray j_c, jint ldc) {
char * transa = (char *)(*env)->GetStringUTFChars(env, j_transa, 0);
char * matdescra = (char *)(*env)->GetStringUTFChars(env, j_matdescra, 0);
jdouble * vals = (*env)->GetPrimitiveArrayCritical(env, j_vals, 0);
jint * ir = (*env)->GetPrimitiveArrayCritical(env, j_ir, 0);
jint * jc = (*env)->GetPrimitiveArrayCritical(env, j_jc, 0);
jdouble * b = (*env)->GetPrimitiveArrayCritical(env, j_b, 0);
jdouble * c = (*env)->GetPrimitiveArrayCritical(env, j_c, 0);
jint returnValue = 0;
if (transa != NULL && matdescra != NULL && vals != NULL && ir != NULL && jc != NULL && b != NULL && c != NULL) {
mkl_dcsrmm(transa, &m, &n, &k, &alpha, matdescra, vals, ir, jc, jc+1, b, &ldb, &beta, c, &ldc);
} else {
returnValue = 1;
}
(*env)->ReleasePrimitiveArrayCritical(env, j_c, c, 0);
(*env)->ReleasePrimitiveArrayCritical(env, j_b, b, 0);
(*env)->ReleasePrimitiveArrayCritical(env, j_jc, jc, 0);
(*env)->ReleasePrimitiveArrayCritical(env, j_ir, ir, 0);
(*env)->ReleasePrimitiveArrayCritical(env, j_vals, vals, 0);
(*env)->ReleaseStringUTFChars(env, j_matdescra, matdescra);
(*env)->ReleaseStringUTFChars(env, j_transa, transa);
return returnValue;
};
void multiply(const CrsMatrix< double , Kokkos::OpenMP >& A,
const std::vector< Kokkos::View< double* , Kokkos::OpenMP > >& x,
std::vector< Kokkos::View< double* , Kokkos::OpenMP > >& y,
MKLMultiply tag)
{
typedef Kokkos::OpenMP device_type ;
typedef double value_type ;
typedef Kokkos::View< double** , Kokkos::LayoutLeft, device_type > trans_multi_vector_type ;
MKL_INT n = A.graph.row_map.dimension_0() - 1 ;
double *A_values = A.values.ptr_on_device() ;
MKL_INT *col_indices = A.graph.entries.ptr_on_device() ;
MKL_INT *row_beg = const_cast<MKL_INT*>(A.graph.row_map.ptr_on_device()) ;
MKL_INT *row_end = row_beg+1;
char matdescra[6] = { 'G', 'x', 'N', 'C', 'x', 'x' };
char trans = 'N';
double alpha = 1.0;
double beta = 0.0;
// Copy columns of x into a contiguous vector
MKL_INT ncol = x.size();
trans_multi_vector_type xx( "xx" , ncol , n );
trans_multi_vector_type yy( "yy" , ncol , n );
Impl::GatherVecTranspose<value_type,device_type>::apply(x,xx);
double *x_values = xx.ptr_on_device() ;
double *y_values = yy.ptr_on_device() ;
// Call MKLs CSR x multi-vector (row-based) multiply
mkl_dcsrmm(&trans, &n, &ncol, &n, &alpha, matdescra, A_values, col_indices,
row_beg, row_end, x_values, &ncol, &beta, y_values, &ncol);
// Copy columns out of continguous multivector
Impl::ScatterVecTranspose<value_type,device_type>::apply(y,yy);
}
void ReducedMassSpringSystemForceModel::GetTangentStiffnessMatrixHelper(
double * tangentStiffnessMatrix)
{
// evaluate stiffness matrix
//PerformanceCounter counter;
massSpringSystem->ComputeStiffnessMatrix(u, sparseMatrix);
//counter.StopCounter();
//printf("counter: %G\n", counter.GetElapsedTime());
// project matrix
#if USE_MKL_SPARSE_BLAS
mkl_set_num_threads(8);
//PerformanceCounter counter;
int upperTriangleOnly=1;
int oneIndexed=1;
sparseMatrix->GenerateCompressedRowMajorFormat_four_array(csr_values, csr_columns, csr_pointerB, csr_pointerE, upperTriangleOnly, oneIndexed);
char transa = 'N';
int m = sparseMatrix->GetNumRows();
int n = r;
int k = m;
double alpha = 1.0;
char matdescra[7] = "SUNFXX";
double * val = csr_values;
int * indx = csr_columns;
int * pntrb = csr_pointerB;
int * pntre = csr_pointerE;
double * b = U;
int ldb = m;
double beta = 0.0;
double * c = bufferMatrix;
int ldc = m;
mkl_dcsrmm(&transa, &m, &n, &k, &alpha, matdescra, val, indx, pntrb, pntre,
b, &ldb, &beta, c, &ldc);
//counter.StopCounter();
//printf("counter: %G\n", counter.GetElapsedTime());
#else
for(int i=0; i<r; i++)
sparseMatrix->MultiplyVector(&U[ELT(3*n,0,i)], &bufferMatrix[ELT(3*n,0,i)]);
#endif
modalMatrix->ProjectMatrix(r, bufferMatrix, tangentStiffnessMatrix);
int r2 = r*r;
for(int i=0; i<r2; i++)
tangentStiffnessMatrix[i] *= -1;
}
