[Olena-patches] 3102: Simplify the normalization of vectors.

* mln/algebra/vec.hh (mln::algebra::vec<n, T>::normalize): Use mln::norm::l2 instead of ad hoc code. --- milena/ChangeLog | 7 +++++++ milena/mln/algebra/vec.hh | 30 +++++++++++++++++++++++------- 2 files changed, 30 insertions(+), 7 deletions(-) diff --git a/milena/ChangeLog b/milena/ChangeLog index 0f3bc29..fae6012 100644 --- a/milena/ChangeLog +++ b/milena/ChangeLog @@ -1,5 +1,12 @@ 2008-12-30 Roland Levillain &lt;roland(a)lrde.epita.fr&gt; + Simplify the normalization of vectors. + + * mln/algebra/vec.hh (mln::algebra::vec<n, T>::normalize): + Use mln::norm::l2 instead of ad hoc code. + +2008-12-30 Roland Levillain &lt;roland(a)lrde.epita.fr&gt; + More complex-related aliases. * mln/core/alias/complex_geometry.hh diff --git a/milena/mln/algebra/vec.hh b/milena/mln/algebra/vec.hh index 2a61693..1434930 100644 --- a/milena/mln/algebra/vec.hh +++ b/milena/mln/algebra/vec.hh @@ -39,6 +39,7 @@ # include <mln/core/concept/object.hh> # include <mln/literal/zero.hh> +# include <mln/norm/l2.hh> # include <mln/trait/all.hh> # include <mln/trait/value_.hh> # include <mln/fun/i2v/all_to.hh> @@ -64,7 +65,10 @@ namespace mln struct zero_t; } - + namespace norm { + template <unsigned n, typename C> + mln_sum(C) l2(const algebra::vec<n,C>& vec); + } namespace trait @@ -211,6 +215,18 @@ namespace mln unsigned size() const; + /* FIXME: What if the vector is null? Even if we choose not to + handle this case, we should *state* in the documentation the + behavior of this method. + + I (Roland) have added an assertion to detect ``erroneous'' + cases, but we might want something different. + ``Implementation defined'' or ``undefined behavior'' is fine + by me, as long as the documentation mentions it. + + FWIW, Trimesh's developers chose to set all the coordinates + of a vector being normalized to 1, when it norm is equal or + lower (sic) to zero. */ const vec<n, T>& normalize(); /// Constructor; coordinates are set by function \p f. @@ -306,7 +322,8 @@ namespace mln operator-(const vec<n,T>& lhs, const vec<n,U>& rhs); // vec * vec - + + /// Scalar product (dot product). template <unsigned n, typename T, typename U> mln_sum_x(T,U) operator*(const vec<n,T>& lhs, const vec<n,U>& rhs); @@ -331,6 +348,7 @@ namespace mln // vprod // FIXME: Generalize... + /// Vectorial product (cross product). template <typename T, typename U> vec<3, mln_trait_op_times(T,U)> // FIXME: Sum of product... vprod(const vec<3, T>& lhs, const vec<3, U>& rhs); @@ -431,12 +449,10 @@ namespace mln inline const vec<n, T>& vec<n, T>::normalize() { - float n_l2 = 0; - for (unsigned i = 0; i < n; ++i) - n_l2 += data_[i] * data_[i]; - n_l2 = float(std::sqrt(n_l2)); + float l2_norm = float(norm::l2(*this)); + mln_assertion(l2_norm > 0.f); for (unsigned i = 0; i < n; ++i) - data_[i] = static_cast<T>(data_[i] / n_l2); + data_[i] = static_cast<T>(data_[i] / l2_norm); return *this; } -- 1.6.0.4