static void run( float a, const Eigen::MatrixBase<Derived2> & A, const Eigen::MatrixBase<Derived1> & x, float b, Eigen::MatrixBase<Derived1> &y) {
EIGEN_STATIC_ASSERT(sizeof(PREC) == sizeof(typename Derived1::Scalar), YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY)
ASSERTMSG(A.cols() == x.rows() && A.rows() == y.rows() ,"ERROR: Vector/Matrix wrong dimension");
// b_dev = alpha * A_dev * x_old_dev + beta *b_dev
//Derived2 C = A;
#if USE_INTEL_BLAS == 1
CBLAS_ORDER order;
CBLAS_TRANSPOSE trans;
if(Derived1::Flags & Eigen::RowMajorBit) {
order = CblasRowMajor;
} else {
order = CblasColMajor;
}
trans = CblasNoTrans;
mkl_set_dynamic(false);
mkl_set_num_threads(BLAS_NUM_THREADS);
//cout << "Threads:" << mkl_get_max_threads();
cblas_sgemv(order, trans, A.rows(), A.cols(), a, const_cast<double*>(&(A.operator()(0,0))), A.outerStride(), const_cast<double*>(&(x.operator()(0,0))), 1, b, &(y.operator()(0,0)), 1);
//cblas_dgemm(order,trans,trans, A.rows(), A.cols(), A.cols(), 1.0, const_cast<double*>(&(A.operator()(0,0))), A.rows(), const_cast<double*>(&(A.operator()(0,0))), A.rows(), 1.0 , &(C.operator()(0,0)), C.rows() );
#else
#if USE_GOTO_BLAS == 1
/* static DGEMVFunc DGEMV = NULL;
if (DGEMV == NULL) {
HINSTANCE hInstLibrary = LoadLibrary("libopenblasp-r0.1alpha2.2.dll");
DGEMV = (DGEMVFunc)GetProcAddress(hInstLibrary, "DGEMV");
}*/
char trans = 'N';
BLAS_INT idx = 1;
BLAS_INT m = A.rows();
BLAS_INT n = A.cols();
sgemv(&trans, &m, &n, &a, &(A.operator()(0,0)), &m, &(x.operator()(0,0)), &idx, &b, &(y.operator()(0,0)), &idx);
// FreeLibrary(hInstLibrary);
#else
ASSERTMSG(false,"No implementation for BLAS defined!");
#endif
#endif
}
