https://svn.lrde.epita.fr/svn/oln/trunk/milena Index: ChangeLog from Thierry Geraud <thierry.geraud@lrde.epita.fr> Add morpho elementary gradient fastest impl for functions. * mln/morpho/elementary/gradient.hh (todo): New. (gradient_on_function_fastest): New impl. (gradient_dispatch): New overload. gradient.hh | 47 +++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 45 insertions(+), 2 deletions(-) Index: mln/morpho/elementary/gradient.hh --- mln/morpho/elementary/gradient.hh (revision 3221) +++ mln/morpho/elementary/gradient.hh (working copy) @@ -31,7 +31,9 @@ /// \file mln/morpho/elementary/gradient.hh /// -/// \todo Add fastest versions (be careful about extension values...) +/// \todo Add fastest version for sets. +/// +/// \todo Replace .domain().has() by .has()! # include <mln/morpho/includes.hh> # include <mln/accu/min_max.hh> @@ -92,7 +94,7 @@ mln_niter(N) n(nbh, p); for_all(p) { - a.init(); + a.take_as_init(input(p)); for_all(n) if (input.domain().has(n)) a.take(input(n)); output(p) = a.second() - a.first(); @@ -142,6 +144,38 @@ return output; } + + template <typename I, typename N> + mln_concrete(I) + gradient_on_function_fastest(const Image<I>& input_, const Neighborhood<N>& nbh_) + { + trace::entering("morpho::elementary::impl::gradient_on_function_fastest"); + + const I& input = exact(input_); + const N& nbh = exact(nbh_); + internal::gradient_tests(input, nbh); + + accu::min_max<mln_value(I)> a; + extension::adjust_duplicate(input, nbh); + + mln_concrete(I) output; + initialize(output, input); + + mln_pixter(const I) p_in(input); + mln_pixter(I) p_out(output); + mln_nixter(const I, N) n(p_in, nbh); + for_all_2(p_in, p_out) + { + a.take_as_init(p_in.val()); + for_all(n) + a.take(n.val()); + p_out.val() = a.second() - a.first(); + } + + trace::exiting("morpho::elementary::impl::gradient_on_function_fastest"); + return output; + } + } // end of namespace mln::morpho::elementary::impl @@ -161,6 +195,15 @@ template <typename I, typename N> mln_concrete(I) + gradient_dispatch(trait::image::kind::any, + trait::image::speed::fastest, + const Image<I>& input, const Neighborhood<N>& nbh) + { + return impl::gradient_on_function_fastest(input, nbh); + } + + template <typename I, typename N> + mln_concrete(I) gradient_dispatch(trait::image::kind::logic, trait::image::speed::any, const Image<I>& input, const Neighborhood<N>& nbh)