OrcaSlicer/src/libslic3r/Point.hpp
salt.wei aab8a12801 ENH: add arachne engine for narrow internal solid infill
ConcentricGapFill pattern was used for internal narrow
solid infill. Use arachne engine instead to remove
gap fill inside the pattern and improve the extrusion path

Signed-off-by: salt.wei <salt.wei@bambulab.com>
Change-Id: I758d7c72eb71cc37026b7cebf746cc345014c3f5
(cherry picked from commit 0b6bacd21a091afc13d7b36a69e5b10f155bc6f8)
2022-08-26 09:25:59 +08:00

561 lines
24 KiB
C++

#ifndef slic3r_Point_hpp_
#define slic3r_Point_hpp_
#include "libslic3r.h"
#include <cstddef>
#include <vector>
#include <cmath>
#include <string>
#include <sstream>
#include <unordered_map>
#include <Eigen/Geometry>
#include "LocalesUtils.hpp"
namespace Slic3r {
class BoundingBox;
class BoundingBoxf;
class Line;
class MultiPoint;
class Point;
using Vector = Point;
// Base template for eigen derived vectors
template<int N, int M, class T>
using Mat = Eigen::Matrix<T, N, M, Eigen::DontAlign, N, M>;
template<int N, class T> using Vec = Mat<N, 1, T>;
// Eigen types, to replace the Slic3r's own types in the future.
// Vector types with a fixed point coordinate base type.
using Vec2crd = Eigen::Matrix<coord_t, 2, 1, Eigen::DontAlign>;
using Vec3crd = Eigen::Matrix<coord_t, 3, 1, Eigen::DontAlign>;
using Vec2i = Eigen::Matrix<int, 2, 1, Eigen::DontAlign>;
using Vec3i = Eigen::Matrix<int, 3, 1, Eigen::DontAlign>;
using Vec4i = Eigen::Matrix<int, 4, 1, Eigen::DontAlign>;
using Vec2i32 = Eigen::Matrix<int32_t, 2, 1, Eigen::DontAlign>;
using Vec2i64 = Eigen::Matrix<int64_t, 2, 1, Eigen::DontAlign>;
using Vec3i32 = Eigen::Matrix<int32_t, 3, 1, Eigen::DontAlign>;
using Vec3i64 = Eigen::Matrix<int64_t, 3, 1, Eigen::DontAlign>;
// Vector types with a double coordinate base type.
using Vec2f = Eigen::Matrix<float, 2, 1, Eigen::DontAlign>;
using Vec3f = Eigen::Matrix<float, 3, 1, Eigen::DontAlign>;
using Vec2d = Eigen::Matrix<double, 2, 1, Eigen::DontAlign>;
using Vec3d = Eigen::Matrix<double, 3, 1, Eigen::DontAlign>;
// BBS
using Vec4d = Eigen::Matrix<double, 4, 1, Eigen::DontAlign>;
using Points = std::vector<Point>;
using PointPtrs = std::vector<Point*>;
using PointConstPtrs = std::vector<const Point*>;
using Points3 = std::vector<Vec3crd>;
using Pointfs = std::vector<Vec2d>;
using Vec2ds = std::vector<Vec2d>;
using Pointf3s = std::vector<Vec3d>;
using Matrix2f = Eigen::Matrix<float, 2, 2, Eigen::DontAlign>;
using Matrix2d = Eigen::Matrix<double, 2, 2, Eigen::DontAlign>;
using Matrix3f = Eigen::Matrix<float, 3, 3, Eigen::DontAlign>;
using Matrix3d = Eigen::Matrix<double, 3, 3, Eigen::DontAlign>;
using Matrix4f = Eigen::Matrix<float, 4, 4, Eigen::DontAlign>;
using Matrix4d = Eigen::Matrix<double, 4, 4, Eigen::DontAlign>;
template<int N, class T>
using Transform = Eigen::Transform<float, N, Eigen::Affine, Eigen::DontAlign>;
using Transform2f = Eigen::Transform<float, 2, Eigen::Affine, Eigen::DontAlign>;
using Transform2d = Eigen::Transform<double, 2, Eigen::Affine, Eigen::DontAlign>;
using Transform3f = Eigen::Transform<float, 3, Eigen::Affine, Eigen::DontAlign>;
using Transform3d = Eigen::Transform<double, 3, Eigen::Affine, Eigen::DontAlign>;
// I don't know why Eigen::Transform::Identity() return a const object...
template<int N, class T> Transform<N, T> identity() { return Transform<N, T>::Identity(); }
inline const auto &identity3f = identity<3, float>;
inline const auto &identity3d = identity<3, double>;
inline bool operator<(const Vec2d &lhs, const Vec2d &rhs) { return lhs.x() < rhs.x() || (lhs.x() == rhs.x() && lhs.y() < rhs.y()); }
template<int Options>
int32_t cross2(const Eigen::MatrixBase<Eigen::Matrix<int32_t, 2, 1, Options>> &v1, const Eigen::MatrixBase<Eigen::Matrix<int32_t, 2, 1, Options>> &v2) = delete;
template<typename T, int Options>
inline T cross2(const Eigen::MatrixBase<Eigen::Matrix<T, 2, 1, Options>> &v1, const Eigen::MatrixBase<Eigen::Matrix<T, 2, 1, Options>> &v2)
{
return v1.x() * v2.y() - v1.y() * v2.x();
}
template<typename Derived, typename Derived2>
inline typename Derived::Scalar cross2(const Eigen::MatrixBase<Derived> &v1, const Eigen::MatrixBase<Derived2> &v2)
{
static_assert(std::is_same<typename Derived::Scalar, typename Derived2::Scalar>::value, "cross2(): Scalar types of 1st and 2nd operand must be equal.");
return v1.x() * v2.y() - v1.y() * v2.x();
}
template<typename T, int Options>
inline Eigen::Matrix<T, 2, 1, Eigen::DontAlign> perp(const Eigen::MatrixBase<Eigen::Matrix<T, 2, 1, Options>> &v) { return Eigen::Matrix<T, 2, 1, Eigen::DontAlign>(- v.y(), v.x()); }
template<class T, int N, int Options>
Eigen::Matrix<T, 2, 1, Eigen::DontAlign> to_2d(const Eigen::MatrixBase<Eigen::Matrix<T, N, 1, Options>> &ptN) { return { ptN.x(), ptN.y() }; }
template<class T, int Options>
Eigen::Matrix<T, 3, 1, Eigen::DontAlign> to_3d(const Eigen::MatrixBase<Eigen::Matrix<T, 2, 1, Options>> & pt, const T z) { return { pt.x(), pt.y(), z }; }
inline Vec2d unscale(coord_t x, coord_t y) { return Vec2d(unscale<double>(x), unscale<double>(y)); }
inline Vec2d unscale(const Vec2crd &pt) { return Vec2d(unscale<double>(pt.x()), unscale<double>(pt.y())); }
inline Vec2d unscale(const Vec2d &pt) { return Vec2d(unscale<double>(pt.x()), unscale<double>(pt.y())); }
inline Vec3d unscale(coord_t x, coord_t y, coord_t z) { return Vec3d(unscale<double>(x), unscale<double>(y), unscale<double>(z)); }
inline Vec3d unscale(const Vec3crd &pt) { return Vec3d(unscale<double>(pt.x()), unscale<double>(pt.y()), unscale<double>(pt.z())); }
inline Vec3d unscale(const Vec3d &pt) { return Vec3d(unscale<double>(pt.x()), unscale<double>(pt.y()), unscale<double>(pt.z())); }
inline std::string to_string(const Vec2crd &pt) { return std::string("[") + float_to_string_decimal_point(pt.x()) + ", " + float_to_string_decimal_point(pt.y()) + "]"; }
inline std::string to_string(const Vec2d &pt) { return std::string("[") + float_to_string_decimal_point(pt.x()) + ", " + float_to_string_decimal_point(pt.y()) + "]"; }
inline std::string to_string(const Vec3crd &pt) { return std::string("[") + float_to_string_decimal_point(pt.x()) + ", " + float_to_string_decimal_point(pt.y()) + ", " + float_to_string_decimal_point(pt.z()) + "]"; }
inline std::string to_string(const Vec3d &pt) { return std::string("[") + float_to_string_decimal_point(pt.x()) + ", " + float_to_string_decimal_point(pt.y()) + ", " + float_to_string_decimal_point(pt.z()) + "]"; }
std::vector<Vec3f> transform(const std::vector<Vec3f>& points, const Transform3f& t);
Pointf3s transform(const Pointf3s& points, const Transform3d& t);
template<int N, class T> using Vec = Eigen::Matrix<T, N, 1, Eigen::DontAlign, N, 1>;
class Point : public Vec2crd
{
public:
using coord_type = coord_t;
Point() : Vec2crd(0, 0) {}
Point(int32_t x, int32_t y) : Vec2crd(coord_t(x), coord_t(y)) {}
Point(int64_t x, int64_t y) : Vec2crd(coord_t(x), coord_t(y)) {}
Point(double x, double y) : Vec2crd(coord_t(lrint(x)), coord_t(lrint(y))) {}
Point(const Point &rhs) { *this = rhs; }
explicit Point(const Vec2d& rhs) : Vec2crd(coord_t(lrint(rhs.x())), coord_t(lrint(rhs.y()))) {}
// This constructor allows you to construct Point from Eigen expressions
template<typename OtherDerived>
Point(const Eigen::MatrixBase<OtherDerived> &other) : Vec2crd(other) {}
static Point new_scale(coordf_t x, coordf_t y) { return Point(coord_t(scale_(x)), coord_t(scale_(y))); }
static Point new_scale(const Vec2d &v) { return Point(coord_t(scale_(v.x())), coord_t(scale_(v.y()))); }
static Point new_scale(const Vec2f &v) { return Point(coord_t(scale_(v.x())), coord_t(scale_(v.y()))); }
// This method allows you to assign Eigen expressions to MyVectorType
template<typename OtherDerived>
Point& operator=(const Eigen::MatrixBase<OtherDerived> &other)
{
this->Vec2crd::operator=(other);
return *this;
}
Point& operator+=(const Point& rhs) { this->x() += rhs.x(); this->y() += rhs.y(); return *this; }
Point& operator-=(const Point& rhs) { this->x() -= rhs.x(); this->y() -= rhs.y(); return *this; }
Point& operator*=(const double &rhs) { this->x() = coord_t(this->x() * rhs); this->y() = coord_t(this->y() * rhs); return *this; }
Point operator*(const double &rhs) { return Point(this->x() * rhs, this->y() * rhs); }
void rotate(double angle) { this->rotate(std::cos(angle), std::sin(angle)); }
void rotate(double cos_a, double sin_a) {
double cur_x = (double)this->x();
double cur_y = (double)this->y();
this->x() = (coord_t)round(cos_a * cur_x - sin_a * cur_y);
this->y() = (coord_t)round(cos_a * cur_y + sin_a * cur_x);
}
void rotate(double angle, const Point &center);
Point rotated(double angle) const { Point res(*this); res.rotate(angle); return res; }
Point rotated(double cos_a, double sin_a) const { Point res(*this); res.rotate(cos_a, sin_a); return res; }
Point rotated(double angle, const Point &center) const { Point res(*this); res.rotate(angle, center); return res; }
Point rotate_90_degree_ccw() const { return Point(-this->y(), this->x()); }
int nearest_point_index(const Points &points) const;
int nearest_point_index(const PointConstPtrs &points) const;
int nearest_point_index(const PointPtrs &points) const;
bool nearest_point(const Points &points, Point* point) const;
double ccw(const Point &p1, const Point &p2) const;
double ccw(const Line &line) const;
double ccw_angle(const Point &p1, const Point &p2) const;
Point projection_onto(const MultiPoint &poly) const;
Point projection_onto(const Line &line) const;
};
inline bool operator<(const Point &l, const Point &r)
{
return l.x() < r.x() || (l.x() == r.x() && l.y() < r.y());
}
inline bool is_approx(const Point &p1, const Point &p2, coord_t epsilon = coord_t(SCALED_EPSILON))
{
Point d = (p2 - p1).cwiseAbs();
return d.x() < epsilon && d.y() < epsilon;
}
inline bool is_approx(const Vec2f &p1, const Vec2f &p2, float epsilon = float(EPSILON))
{
Vec2f d = (p2 - p1).cwiseAbs();
return d.x() < epsilon && d.y() < epsilon;
}
inline bool is_approx(const Vec2d &p1, const Vec2d &p2, double epsilon = EPSILON)
{
Vec2d d = (p2 - p1).cwiseAbs();
return d.x() < epsilon && d.y() < epsilon;
}
inline bool is_approx(const Vec3f &p1, const Vec3f &p2, float epsilon = float(EPSILON))
{
Vec3f d = (p2 - p1).cwiseAbs();
return d.x() < epsilon && d.y() < epsilon && d.z() < epsilon;
}
inline bool is_approx(const Vec3d &p1, const Vec3d &p2, double epsilon = EPSILON)
{
Vec3d d = (p2 - p1).cwiseAbs();
return d.x() < epsilon && d.y() < epsilon && d.z() < epsilon;
}
inline Point lerp(const Point &a, const Point &b, double t)
{
assert((t >= -EPSILON) && (t <= 1. + EPSILON));
return ((1. - t) * a.cast<double>() + t * b.cast<double>()).cast<coord_t>();
}
BoundingBox get_extents(const Points &pts);
BoundingBox get_extents(const std::vector<Points> &pts);
BoundingBoxf get_extents(const std::vector<Vec2d> &pts);
// Test for duplicate points in a vector of points.
// The points are copied, sorted and checked for duplicates globally.
bool has_duplicate_points(std::vector<Point> &&pts);
inline bool has_duplicate_points(const std::vector<Point> &pts)
{
std::vector<Point> cpy = pts;
return has_duplicate_points(std::move(cpy));
}
// Test for duplicate points in a vector of points.
// Only successive points are checked for equality.
inline bool has_duplicate_successive_points(const std::vector<Point> &pts)
{
for (size_t i = 1; i < pts.size(); ++ i)
if (pts[i - 1] == pts[i])
return true;
return false;
}
// Test for duplicate points in a vector of points.
// Only successive points are checked for equality. Additionally, first and last points are compared for equality.
inline bool has_duplicate_successive_points_closed(const std::vector<Point> &pts)
{
return has_duplicate_successive_points(pts) || (pts.size() >= 2 && pts.front() == pts.back());
}
inline bool shorter_then(const Point& p0, const coord_t len)
{
if (p0.x() > len || p0.x() < -len)
return false;
if (p0.y() > len || p0.y() < -len)
return false;
return p0.cast<int64_t>().squaredNorm() <= Slic3r::sqr(int64_t(len));
}
namespace int128 {
// Exact orientation predicate,
// returns +1: CCW, 0: collinear, -1: CW.
int orient(const Vec2crd &p1, const Vec2crd &p2, const Vec2crd &p3);
// Exact orientation predicate,
// returns +1: CCW, 0: collinear, -1: CW.
int cross(const Vec2crd &v1, const Vec2crd &v2);
}
// To be used by std::unordered_map, std::unordered_multimap and friends.
struct PointHash {
size_t operator()(const Vec2crd &pt) const {
return coord_t((89 * 31 + int64_t(pt.x())) * 31 + pt.y());
}
};
// A generic class to search for a closest Point in a given radius.
// It uses std::unordered_multimap to implement an efficient 2D spatial hashing.
// The PointAccessor has to return const Point*.
// If a nullptr is returned, it is ignored by the query.
template<typename ValueType, typename PointAccessor> class ClosestPointInRadiusLookup
{
public:
ClosestPointInRadiusLookup(coord_t search_radius, PointAccessor point_accessor = PointAccessor()) :
m_search_radius(search_radius), m_point_accessor(point_accessor), m_grid_log2(0)
{
// Resolution of a grid, twice the search radius + some epsilon.
coord_t gridres = 2 * m_search_radius + 4;
m_grid_resolution = gridres;
assert(m_grid_resolution > 0);
assert(m_grid_resolution < (coord_t(1) << 30));
// Compute m_grid_log2 = log2(m_grid_resolution)
if (m_grid_resolution > 32767) {
m_grid_resolution >>= 16;
m_grid_log2 += 16;
}
if (m_grid_resolution > 127) {
m_grid_resolution >>= 8;
m_grid_log2 += 8;
}
if (m_grid_resolution > 7) {
m_grid_resolution >>= 4;
m_grid_log2 += 4;
}
if (m_grid_resolution > 1) {
m_grid_resolution >>= 2;
m_grid_log2 += 2;
}
if (m_grid_resolution > 0)
++ m_grid_log2;
m_grid_resolution = 1 << m_grid_log2;
assert(m_grid_resolution >= gridres);
assert(gridres > m_grid_resolution / 2);
}
void insert(const ValueType &value) {
const Vec2crd *pt = m_point_accessor(value);
if (pt != nullptr)
m_map.emplace(std::make_pair(Vec2crd(pt->x()>>m_grid_log2, pt->y()>>m_grid_log2), value));
}
void insert(ValueType &&value) {
const Vec2crd *pt = m_point_accessor(value);
if (pt != nullptr)
m_map.emplace(std::make_pair(Vec2crd(pt->x()>>m_grid_log2, pt->y()>>m_grid_log2), std::move(value)));
}
// Erase a data point equal to value. (ValueType has to declare the operator==).
// Returns true if the data point equal to value was found and removed.
bool erase(const ValueType &value) {
const Point *pt = m_point_accessor(value);
if (pt != nullptr) {
// Range of fragment starts around grid_corner, close to pt.
auto range = m_map.equal_range(Point((*pt).x()>>m_grid_log2, (*pt).y()>>m_grid_log2));
// Remove the first item.
for (auto it = range.first; it != range.second; ++ it) {
if (it->second == value) {
m_map.erase(it);
return true;
}
}
}
return false;
}
// Return a pair of <ValueType*, distance_squared>
std::pair<const ValueType*, double> find(const Vec2crd &pt) {
// Iterate over 4 closest grid cells around pt,
// find the closest start point inside these cells to pt.
const ValueType *value_min = nullptr;
double dist_min = std::numeric_limits<double>::max();
// Round pt to a closest grid_cell corner.
Vec2crd grid_corner((pt.x()+(m_grid_resolution>>1))>>m_grid_log2, (pt.y()+(m_grid_resolution>>1))>>m_grid_log2);
// For four neighbors of grid_corner:
for (coord_t neighbor_y = -1; neighbor_y < 1; ++ neighbor_y) {
for (coord_t neighbor_x = -1; neighbor_x < 1; ++ neighbor_x) {
// Range of fragment starts around grid_corner, close to pt.
auto range = m_map.equal_range(Vec2crd(grid_corner.x() + neighbor_x, grid_corner.y() + neighbor_y));
// Find the map entry closest to pt.
for (auto it = range.first; it != range.second; ++it) {
const ValueType &value = it->second;
const Vec2crd *pt2 = m_point_accessor(value);
if (pt2 != nullptr) {
const double d2 = (pt - *pt2).cast<double>().squaredNorm();
if (d2 < dist_min) {
dist_min = d2;
value_min = &value;
}
}
}
}
}
return (value_min != nullptr && dist_min < coordf_t(m_search_radius) * coordf_t(m_search_radius)) ?
std::make_pair(value_min, dist_min) :
std::make_pair(nullptr, std::numeric_limits<double>::max());
}
// Returns all pairs of values and squared distances.
std::vector<std::pair<const ValueType*, double>> find_all(const Vec2crd &pt) {
// Iterate over 4 closest grid cells around pt,
// Round pt to a closest grid_cell corner.
Vec2crd grid_corner((pt.x()+(m_grid_resolution>>1))>>m_grid_log2, (pt.y()+(m_grid_resolution>>1))>>m_grid_log2);
// For four neighbors of grid_corner:
std::vector<std::pair<const ValueType*, double>> out;
const double r2 = double(m_search_radius) * m_search_radius;
for (coord_t neighbor_y = -1; neighbor_y < 1; ++ neighbor_y) {
for (coord_t neighbor_x = -1; neighbor_x < 1; ++ neighbor_x) {
// Range of fragment starts around grid_corner, close to pt.
auto range = m_map.equal_range(Vec2crd(grid_corner.x() + neighbor_x, grid_corner.y() + neighbor_y));
// Find the map entry closest to pt.
for (auto it = range.first; it != range.second; ++it) {
const ValueType &value = it->second;
const Vec2crd *pt2 = m_point_accessor(value);
if (pt2 != nullptr) {
const double d2 = (pt - *pt2).cast<double>().squaredNorm();
if (d2 <= r2)
out.emplace_back(&value, d2);
}
}
}
}
return out;
}
private:
using map_type = typename std::unordered_multimap<Vec2crd, ValueType, PointHash>;
PointAccessor m_point_accessor;
map_type m_map;
coord_t m_search_radius;
coord_t m_grid_resolution;
coord_t m_grid_log2;
};
std::ostream& operator<<(std::ostream &stm, const Vec2d &pointf);
// /////////////////////////////////////////////////////////////////////////////
// Type safe conversions to and from scaled and unscaled coordinates
// /////////////////////////////////////////////////////////////////////////////
// Semantics are the following:
// Upscaling (scaled()): only from floating point types (or Vec) to either
// floating point or integer 'scaled coord' coordinates.
// Downscaling (unscaled()): from arithmetic (or Vec) to floating point only
// Conversion definition from unscaled to floating point scaled
template<class Tout,
class Tin,
class = FloatingOnly<Tin>>
inline constexpr FloatingOnly<Tout> scaled(const Tin &v) noexcept
{
return Tout(v / Tin(SCALING_FACTOR));
}
// Conversion definition from unscaled to integer 'scaled coord'.
// TODO: is the rounding necessary? Here it is commented out to show that
// it can be different for integers but it does not have to be. Using
// std::round means loosing noexcept and constexpr modifiers
template<class Tout = coord_t, class Tin, class = FloatingOnly<Tin>>
inline constexpr ScaledCoordOnly<Tout> scaled(const Tin &v) noexcept
{
//return static_cast<Tout>(std::round(v / SCALING_FACTOR));
return Tout(v / Tin(SCALING_FACTOR));
}
// Conversion for Eigen vectors (N dimensional points)
template<class Tout = coord_t,
class Tin,
int N,
class = FloatingOnly<Tin>,
int...EigenArgs>
inline Eigen::Matrix<ArithmeticOnly<Tout>, N, EigenArgs...>
scaled(const Eigen::Matrix<Tin, N, EigenArgs...> &v)
{
return (v / SCALING_FACTOR).template cast<Tout>();
}
// Conversion from arithmetic scaled type to floating point unscaled
template<class Tout = double,
class Tin,
class = ArithmeticOnly<Tin>,
class = FloatingOnly<Tout>>
inline constexpr Tout unscaled(const Tin &v) noexcept
{
return Tout(v) * Tout(SCALING_FACTOR);
}
// Unscaling for Eigen vectors. Input base type can be arithmetic, output base
// type can only be floating point.
template<class Tout = double,
class Tin,
int N,
class = ArithmeticOnly<Tin>,
class = FloatingOnly<Tout>,
int...EigenArgs>
inline constexpr Eigen::Matrix<Tout, N, EigenArgs...>
unscaled(const Eigen::Matrix<Tin, N, EigenArgs...> &v) noexcept
{
return v.template cast<Tout>() * Tout(SCALING_FACTOR);
}
// Align a coordinate to a grid. The coordinate may be negative,
// the aligned value will never be bigger than the original one.
inline coord_t align_to_grid(const coord_t coord, const coord_t spacing) {
// Current C++ standard defines the result of integer division to be rounded to zero,
// for both positive and negative numbers. Here we want to round down for negative
// numbers as well.
coord_t aligned = (coord < 0) ?
((coord - spacing + 1) / spacing) * spacing :
(coord / spacing) * spacing;
assert(aligned <= coord);
return aligned;
}
inline Point align_to_grid(Point coord, Point spacing)
{ return Point(align_to_grid(coord.x(), spacing.x()), align_to_grid(coord.y(), spacing.y())); }
inline coord_t align_to_grid(coord_t coord, coord_t spacing, coord_t base)
{ return base + align_to_grid(coord - base, spacing); }
inline Point align_to_grid(Point coord, Point spacing, Point base)
{ return Point(align_to_grid(coord.x(), spacing.x(), base.x()), align_to_grid(coord.y(), spacing.y(), base.y())); }
} // namespace Slic3r
// start Boost
#include <boost/version.hpp>
#include <boost/polygon/polygon.hpp>
namespace boost { namespace polygon {
template <>
struct geometry_concept<Slic3r::Point> { using type = point_concept; };
template <>
struct point_traits<Slic3r::Point> {
using coordinate_type = coord_t;
static inline coordinate_type get(const Slic3r::Point& point, orientation_2d orient) {
return static_cast<coordinate_type>(point((orient == HORIZONTAL) ? 0 : 1));
}
};
template <>
struct point_mutable_traits<Slic3r::Point> {
using coordinate_type = coord_t;
static inline void set(Slic3r::Point& point, orientation_2d orient, coord_t value) {
point((orient == HORIZONTAL) ? 0 : 1) = value;
}
static inline Slic3r::Point construct(coord_t x_value, coord_t y_value) {
return Slic3r::Point(x_value, y_value);
}
};
} }
// end Boost
// Serialization through the Cereal library
namespace cereal {
// template<class Archive> void serialize(Archive& archive, Slic3r::Vec2crd &v) { archive(v.x(), v.y()); }
// template<class Archive> void serialize(Archive& archive, Slic3r::Vec3crd &v) { archive(v.x(), v.y(), v.z()); }
template<class Archive> void serialize(Archive& archive, Slic3r::Vec2i &v) { archive(v.x(), v.y()); }
template<class Archive> void serialize(Archive& archive, Slic3r::Vec3i &v) { archive(v.x(), v.y(), v.z()); }
// template<class Archive> void serialize(Archive& archive, Slic3r::Vec2i64 &v) { archive(v.x(), v.y()); }
// template<class Archive> void serialize(Archive& archive, Slic3r::Vec3i64 &v) { archive(v.x(), v.y(), v.z()); }
template<class Archive> void serialize(Archive& archive, Slic3r::Vec2f &v) { archive(v.x(), v.y()); }
template<class Archive> void serialize(Archive& archive, Slic3r::Vec3f &v) { archive(v.x(), v.y(), v.z()); }
template<class Archive> void serialize(Archive& archive, Slic3r::Vec2d &v) { archive(v.x(), v.y()); }
template<class Archive> void serialize(Archive& archive, Slic3r::Vec3d &v) { archive(v.x(), v.y(), v.z()); }
template<class Archive> void load(Archive& archive, Slic3r::Matrix2f &m) { archive.loadBinary((char*)m.data(), sizeof(float) * 4); }
template<class Archive> void save(Archive& archive, Slic3r::Matrix2f &m) { archive.saveBinary((char*)m.data(), sizeof(float) * 4); }
}
// To be able to use Vec<> and Mat<> in range based for loops:
namespace Eigen {
template<class T, int N, int M>
T* begin(Slic3r::Mat<N, M, T> &mat) { return mat.data(); }
template<class T, int N, int M>
T* end(Slic3r::Mat<N, M, T> &mat) { return mat.data() + N * M; }
template<class T, int N, int M>
const T* begin(const Slic3r::Mat<N, M, T> &mat) { return mat.data(); }
template<class T, int N, int M>
const T* end(const Slic3r::Mat<N, M, T> &mat) { return mat.data() + N * M; }
} // namespace Eigen
#endif