boost/gil/locator.hpp
// // Copyright 2005-2007 Adobe Systems Incorporated // // Distributed under the Boost Software License, Version 1.0 // See accompanying file LICENSE_1_0.txt or copy at // http://www.boost.org/LICENSE_1_0.txt // #ifndef BOOST_GIL_LOCATOR_HPP #define BOOST_GIL_LOCATOR_HPP #include <boost/gil/dynamic_step.hpp> #include <boost/gil/pixel_iterator.hpp> #include <boost/gil/point.hpp> #include <boost/assert.hpp> #include <cstddef> namespace boost { namespace gil { /// Pixel 2D locator //forward declarations template <typename P> std::ptrdiff_t memunit_step(const P*); template <typename P> P* memunit_advanced(const P* p, std::ptrdiff_t diff); template <typename P> P& memunit_advanced_ref(P* p, std::ptrdiff_t diff); template <typename Iterator, typename D> struct iterator_add_deref; template <typename T> class point; namespace detail { // helper class specialized for each axis of pixel_2d_locator template <std::size_t D, typename Loc> class locator_axis; } template <typename T> struct channel_type; template <typename T> struct color_space_type; template <typename T> struct channel_mapping_type; template <typename T> struct is_planar; template <typename T> struct num_channels; /// Base template for types that model HasTransposedTypeConcept. /// The type of a locator or a view that has X and Y swapped. /// By default it is the same. template <typename LocatorOrView> struct transposed_type { using type = LocatorOrView; }; /// \class pixel_2d_locator_base /// \brief base class for models of PixelLocatorConcept /// \ingroup PixelLocatorModel PixelBasedModel /// /// Pixel locator is similar to a pixel iterator, but allows for 2D navigation of pixels within an image view. /// It has a 2D difference_type and supports random access operations like: /// \code /// difference_type offset2(2,3); /// locator+=offset2; /// locator[offset2]=my_pixel; /// \endcode /// /// In addition, each coordinate acts as a random-access iterator that can be modified separately: /// "++locator.x()" or "locator.y()+=10" thereby moving the locator horizontally or vertically. /// /// It is called a locator because it doesn't implement the complete interface of a random access iterator. /// For example, increment and decrement operations don't make sense (no way to specify dimension). /// Also 2D difference between two locators cannot be computed without knowledge of the X position within the image. /// /// This base class provides most of the methods and type aliases needed to create a model of a locator. GIL provides two /// locator models as subclasses of \p pixel_2d_locator_base. A memory-based locator, \p memory_based_2d_locator and a virtual /// locator, \p virtual_2d_locator. /// The minimum functionality a subclass must provide is this: /// \code /// class my_locator : public pixel_2d_locator_base<my_locator, ..., ...> { // supply the types for x-iterator and y-iterator /// using const_t = ...; // read-only locator /// /// template <typename Deref> struct add_deref { /// using type = ...; // locator that invokes the Deref dereference object upon pixel access /// static type make(const my_locator& loc, const Deref& d); /// }; /// /// my_locator(); /// my_locator(const my_locator& pl); /// /// // constructors with dynamic step in y (and x). Only valid for locators with dynamic steps /// my_locator(const my_locator& loc, coord_t y_step); /// my_locator(const my_locator& loc, coord_t x_step, coord_t y_step, bool transpose); /// /// bool operator==(const my_locator& p) const; /// /// // return _references_ to horizontal/vertical iterators. Advancing them moves this locator /// x_iterator& x(); /// y_iterator& y(); /// x_iterator const& x() const; /// y_iterator const& y() const; /// /// // return the vertical distance to another locator. Some models need the horizontal distance to compute it /// y_coord_t y_distance_to(const my_locator& loc2, x_coord_t xDiff) const; /// /// // return true iff incrementing an x-iterator located at the last column will position it at the first /// // column of the next row. Some models need the image width to determine that. /// bool is_1d_traversable(x_coord_t width) const; /// }; /// \endcode /// /// Models may choose to override some of the functions in the base class with more efficient versions. /// template <typename Loc, typename XIterator, typename YIterator> // The concrete subclass, the X-iterator and the Y-iterator class pixel_2d_locator_base { public: using x_iterator = XIterator; using y_iterator = YIterator; // aliasesrequired by ConstRandomAccessNDLocatorConcept static const std::size_t num_dimensions=2; using value_type = typename std::iterator_traits<x_iterator>::value_type; using reference = typename std::iterator_traits<x_iterator>::reference; // result of dereferencing using coord_t = typename std::iterator_traits<x_iterator>::difference_type; // 1D difference type (same for all dimensions) using difference_type = point<coord_t>; // result of operator-(locator,locator) using point_t = difference_type; template <std::size_t D> struct axis { using coord_t = typename detail::locator_axis<D,Loc>::coord_t; using iterator = typename detail::locator_axis<D,Loc>::iterator; }; // aliases required by ConstRandomAccess2DLocatorConcept using x_coord_t = typename point_t::template axis<0>::coord_t; using y_coord_t = typename point_t::template axis<1>::coord_t; bool operator!=(const Loc& p) const { return !(concrete()==p); } x_iterator x_at(x_coord_t dx, y_coord_t dy) const { Loc tmp=concrete(); tmp+=point_t(dx,dy); return tmp.x(); } x_iterator x_at(const difference_type& d) const { Loc tmp=concrete(); tmp+=d; return tmp.x(); } y_iterator y_at(x_coord_t dx, y_coord_t dy) const { Loc tmp=concrete(); tmp+=point_t(dx,dy); return tmp.y(); } y_iterator y_at(const difference_type& d) const { Loc tmp=concrete(); tmp+=d; return tmp.y(); } Loc xy_at(x_coord_t dx, y_coord_t dy) const { Loc tmp=concrete(); tmp+=point_t(dx,dy); return tmp; } Loc xy_at(const difference_type& d) const { Loc tmp=concrete(); tmp+=d; return tmp; } template <std::size_t D> typename axis<D>::iterator& axis_iterator() { return detail::locator_axis<D,Loc>()(concrete()); } template <std::size_t D> typename axis<D>::iterator const& axis_iterator() const { return detail::locator_axis<D,Loc>()(concrete()); } template <std::size_t D> typename axis<D>::iterator axis_iterator(const point_t& p) const { return detail::locator_axis<D,Loc>()(concrete(),p); } reference operator()(x_coord_t dx, y_coord_t dy) const { return *x_at(dx,dy); } reference operator[](const difference_type& d) const { return *x_at(d.x,d.y); } reference operator*() const { return *concrete().x(); } Loc& operator+=(const difference_type& d) { concrete().x()+=d.x; concrete().y()+=d.y; return concrete(); } Loc& operator-=(const difference_type& d) { concrete().x()-=d.x; concrete().y()-=d.y; return concrete(); } Loc operator+(const difference_type& d) const { return xy_at(d); } Loc operator-(const difference_type& d) const { return xy_at(-d); } // Some locators can cache 2D coordinates for faster subsequent access. By default there is no caching using cached_location_t = difference_type; cached_location_t cache_location(const difference_type& d) const { return d; } cached_location_t cache_location(x_coord_t dx, y_coord_t dy)const { return difference_type(dx,dy); } private: Loc& concrete() { return (Loc&)*this; } const Loc& concrete() const { return (const Loc&)*this; } template <typename X> friend class pixel_2d_locator; }; // helper classes for each axis of pixel_2d_locator_base namespace detail { template <typename Loc> class locator_axis<0,Loc> { using point_t = typename Loc::point_t; public: using coord_t = typename point_t::template axis<0>::coord_t; using iterator = typename Loc::x_iterator; inline iterator& operator()( Loc& loc) const { return loc.x(); } inline iterator const& operator()(const Loc& loc) const { return loc.x(); } inline iterator operator()( Loc& loc, const point_t& d) const { return loc.x_at(d); } inline iterator operator()(const Loc& loc, const point_t& d) const { return loc.x_at(d); } }; template <typename Loc> class locator_axis<1,Loc> { using point_t = typename Loc::point_t; public: using coord_t = typename point_t::template axis<1>::coord_t; using iterator = typename Loc::y_iterator; inline iterator& operator()( Loc& loc) const { return loc.y(); } inline iterator const& operator()(const Loc& loc) const { return loc.y(); } inline iterator operator()( Loc& loc, const point_t& d) const { return loc.y_at(d); } inline iterator operator()(const Loc& loc, const point_t& d) const { return loc.y_at(d); } }; } template <typename Loc, typename XIt, typename YIt> struct channel_type<pixel_2d_locator_base<Loc,XIt,YIt> > : public channel_type<XIt> {}; template <typename Loc, typename XIt, typename YIt> struct color_space_type<pixel_2d_locator_base<Loc,XIt,YIt> > : public color_space_type<XIt> {}; template <typename Loc, typename XIt, typename YIt> struct channel_mapping_type<pixel_2d_locator_base<Loc,XIt,YIt> > : public channel_mapping_type<XIt> {}; template <typename Loc, typename XIt, typename YIt> struct is_planar<pixel_2d_locator_base<Loc,XIt,YIt> > : public is_planar<XIt> {}; /// \class memory_based_2d_locator /// \brief Memory-based pixel locator. Models: PixelLocatorConcept,HasDynamicXStepTypeConcept,HasDynamicYStepTypeConcept,HasTransposedTypeConcept /// \ingroup PixelLocatorModel PixelBasedModel /// /// The class takes a step iterator as a parameter. The step iterator provides navigation along the vertical axis /// while its base iterator provides horizontal navigation. /// /// Each instantiation is optimal in terms of size and efficiency. /// For example, xy locator over interleaved rgb image results in a step iterator consisting of /// one std::ptrdiff_t for the row size and one native pointer (8 bytes total). ++locator.x() resolves to pointer /// increment. At the other extreme, a 2D navigation of the even pixels of a planar CMYK image results in a step /// iterator consisting of one std::ptrdiff_t for the doubled row size, and one step iterator consisting of /// one std::ptrdiff_t for the horizontal step of two and a CMYK planar_pixel_iterator consisting of 4 pointers (24 bytes). /// In this case ++locator.x() results in four native pointer additions. /// /// Note also that \p memory_based_2d_locator does not require that its element type be a pixel. It could be /// instantiated with an iterator whose \p value_type models only \p Regular. In this case the locator /// models the weaker RandomAccess2DLocatorConcept, and does not model PixelBasedConcept. /// Many generic algorithms don't require the elements to be pixels. //////////////////////////////////////////////////////////////////////////////////////// template <typename StepIterator> class memory_based_2d_locator : public pixel_2d_locator_base<memory_based_2d_locator<StepIterator>, typename iterator_adaptor_get_base<StepIterator>::type, StepIterator> { using this_t = memory_based_2d_locator<StepIterator>; BOOST_GIL_CLASS_REQUIRE(StepIterator, boost::gil, StepIteratorConcept) public: using parent_t = pixel_2d_locator_base<memory_based_2d_locator<StepIterator>, typename iterator_adaptor_get_base<StepIterator>::type, StepIterator>; using const_t = memory_based_2d_locator<typename const_iterator_type<StepIterator>::type>; // same as this type, but over const values using coord_t = typename parent_t::coord_t; using x_coord_t = typename parent_t::x_coord_t; using y_coord_t = typename parent_t::y_coord_t; using x_iterator = typename parent_t::x_iterator; using y_iterator = typename parent_t::y_iterator; using difference_type = typename parent_t::difference_type; using reference = typename parent_t::reference; template <typename Deref> struct add_deref { using type = memory_based_2d_locator<typename iterator_add_deref<StepIterator,Deref>::type>; static type make(const memory_based_2d_locator<StepIterator>& loc, const Deref& nderef) { return type(iterator_add_deref<StepIterator,Deref>::make(loc.y(),nderef)); } }; memory_based_2d_locator() {} memory_based_2d_locator(const StepIterator& yit) : _p(yit) {} template <typename SI> memory_based_2d_locator(const memory_based_2d_locator<SI>& loc, coord_t y_step) : _p(loc.x(), loc.row_size()*y_step) {} template <typename SI> memory_based_2d_locator(const memory_based_2d_locator<SI>& loc, coord_t x_step, coord_t y_step, bool transpose=false) : _p(make_step_iterator(loc.x(),(transpose ? loc.row_size() : loc.pixel_size())*x_step), (transpose ? loc.pixel_size() : loc.row_size())*y_step ) {} memory_based_2d_locator(x_iterator xit, std::ptrdiff_t row_bytes) : _p(xit,row_bytes) {} template <typename X> memory_based_2d_locator(const memory_based_2d_locator<X>& pl) : _p(pl._p) {} memory_based_2d_locator(const memory_based_2d_locator& pl) : _p(pl._p) {} memory_based_2d_locator& operator=(memory_based_2d_locator const& other) = default; bool operator==(const this_t& p) const { return _p==p._p; } x_iterator const& x() const { return _p.base(); } y_iterator const& y() const { return _p; } x_iterator& x() { return _p.base(); } y_iterator& y() { return _p; } // These are faster versions of functions already provided in the superclass x_iterator x_at (x_coord_t dx, y_coord_t dy) const { return memunit_advanced(x(), offset(dx,dy)); } x_iterator x_at (const difference_type& d) const { return memunit_advanced(x(), offset(d.x,d.y)); } this_t xy_at (x_coord_t dx, y_coord_t dy) const { return this_t(x_at( dx , dy ), row_size()); } this_t xy_at (const difference_type& d) const { return this_t(x_at( d.x, d.y), row_size()); } reference operator()(x_coord_t dx, y_coord_t dy) const { return memunit_advanced_ref(x(),offset(dx,dy)); } reference operator[](const difference_type& d) const { return memunit_advanced_ref(x(),offset(d.x,d.y)); } this_t& operator+=(const difference_type& d) { memunit_advance(x(),offset(d.x,d.y)); return *this; } this_t& operator-=(const difference_type& d) { memunit_advance(x(),offset(-d.x,-d.y)); return *this; } // Memory-based locators can have 1D caching of 2D relative coordinates using cached_location_t = std::ptrdiff_t; // type used to store relative location (to allow for more efficient repeated access) cached_location_t cache_location(const difference_type& d) const { return offset(d.x,d.y); } cached_location_t cache_location(x_coord_t dx, y_coord_t dy)const { return offset(dx,dy); } reference operator[](const cached_location_t& loc) const { return memunit_advanced_ref(x(),loc); } // Only make sense for memory-based locators std::ptrdiff_t row_size() const { return memunit_step(y()); } // distance in mem units (bytes or bits) between adjacent rows std::ptrdiff_t pixel_size() const { return memunit_step(x()); } // distance in mem units (bytes or bits) between adjacent pixels on the same row bool is_1d_traversable(x_coord_t width) const { return row_size()-pixel_size()*width==0; } // is there no gap at the end of each row? // Returns the vertical distance (it2.y-it1.y) between two x_iterators given the difference of their x positions std::ptrdiff_t y_distance_to(this_t const& p2, x_coord_t xDiff) const { std::ptrdiff_t rowDiff = memunit_distance(x(), p2.x()) - pixel_size() * xDiff; BOOST_ASSERT((rowDiff % row_size()) == 0); return rowDiff / row_size(); } private: template <typename X> friend class memory_based_2d_locator; std::ptrdiff_t offset(x_coord_t x, y_coord_t y) const { return y*row_size() + x*pixel_size(); } StepIterator _p; }; ///////////////////////////// // PixelBasedConcept ///////////////////////////// template <typename SI> struct color_space_type<memory_based_2d_locator<SI> > : public color_space_type<typename memory_based_2d_locator<SI>::parent_t> { }; template <typename SI> struct channel_mapping_type<memory_based_2d_locator<SI> > : public channel_mapping_type<typename memory_based_2d_locator<SI>::parent_t> { }; template <typename SI> struct is_planar<memory_based_2d_locator<SI> > : public is_planar<typename memory_based_2d_locator<SI>::parent_t> { }; template <typename SI> struct channel_type<memory_based_2d_locator<SI> > : public channel_type<typename memory_based_2d_locator<SI>::parent_t> { }; ///////////////////////////// // HasDynamicXStepTypeConcept ///////////////////////////// // Take the base iterator of SI (which is typically a step iterator) and change it to have a step in x template <typename SI> struct dynamic_x_step_type<memory_based_2d_locator<SI> > { private: using base_iterator_t = typename iterator_adaptor_get_base<SI>::type; using base_iterator_step_t = typename dynamic_x_step_type<base_iterator_t>::type; using dynamic_step_base_t = typename iterator_adaptor_rebind<SI, base_iterator_step_t>::type; public: using type = memory_based_2d_locator<dynamic_step_base_t>; }; ///////////////////////////// // HasDynamicYStepTypeConcept ///////////////////////////// template <typename SI> struct dynamic_y_step_type<memory_based_2d_locator<SI> > { using type = memory_based_2d_locator<SI>; }; } } // namespace boost::gil #endif