* i.e. multiply all components by
* <tt>factor</tt>.
*/
- Tensor<rank_,dim> & operator *= (const double &factor);
+ Tensor<rank_,dim> & operator *= (const double factor);
/**
* Scale the vector by
* <tt>1/factor</tt>.
*/
- Tensor<rank_,dim> & operator /= (const double &factor);
+ Tensor<rank_,dim> & operator /= (const double factor);
/**
* Add two tensors. If possible, you
template <int rank_, int dim>
inline
Tensor<rank_,dim> &
-Tensor<rank_,dim>::operator *= (const double &s)
+Tensor<rank_,dim>::operator *= (const double s)
{
for (unsigned int i=0; i<dim; ++i)
subtensor[i] *= s;
template <int rank_, int dim>
inline
Tensor<rank_,dim> &
-Tensor<rank_,dim>::operator /= (const double &s)
+Tensor<rank_,dim>::operator /= (const double s)
{
for (unsigned int i=0; i<dim; ++i)
subtensor[i] /= s;
// $Id$
// Version: $Name$
//
-// Copyright (C) 1998, 1999, 2000, 2001, 2002, 2003, 2004 by the deal authors
+// Copyright (C) 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005 by the deal authors
//
// This file is subject to QPL and may not be distributed
// without copyright and license information. Please refer
* <tt>factor</tt>, i.e. multiply all
* coordinates by <tt>factor</tt>.
*/
- Tensor<1,dim> & operator *= (const double &factor);
+ Tensor<1,dim> & operator *= (const double factor);
/**
* Scale the vector by <tt>1/factor</tt>.
*/
- Tensor<1,dim> & operator /= (const double &factor);
+ Tensor<1,dim> & operator /= (const double factor);
/**
* Returns the scalar product of
template <int dim>
inline
-Tensor<1,dim> & Tensor<1,dim>::operator *= (const double &s)
+Tensor<1,dim> & Tensor<1,dim>::operator *= (const double s)
{
for (unsigned int i=0; i<dim; ++i)
values[i] *= s;
template <int dim>
inline
-Tensor<1,dim> & Tensor<1,dim>::operator /= (const double &s)
+Tensor<1,dim> & Tensor<1,dim>::operator /= (const double s)
{
for (unsigned int i=0; i<dim; ++i)
values[i] /= s;