forked from Mirrorlandia_minetest/irrlicht
490 lines
17 KiB
C++
490 lines
17 KiB
C++
// Copyright (C) 2002-2012 Nikolaus Gebhardt
|
|
// This file is part of the "Irrlicht Engine".
|
|
// For conditions of distribution and use, see copyright notice in irrlicht.h
|
|
|
|
#ifndef __IRR_POINT_3D_H_INCLUDED__
|
|
#define __IRR_POINT_3D_H_INCLUDED__
|
|
|
|
#include "irrMath.h"
|
|
|
|
#include <functional>
|
|
|
|
namespace irr
|
|
{
|
|
namespace core
|
|
{
|
|
|
|
//! 3d vector template class with lots of operators and methods.
|
|
/** The vector3d class is used in Irrlicht for three main purposes:
|
|
1) As a direction vector (most of the methods assume this).
|
|
2) As a position in 3d space (which is synonymous with a direction vector from the origin to this position).
|
|
3) To hold three Euler rotations, where X is pitch, Y is yaw and Z is roll.
|
|
*/
|
|
template <class T>
|
|
class vector3d
|
|
{
|
|
public:
|
|
//! Default constructor (null vector).
|
|
vector3d() : X(0), Y(0), Z(0) {}
|
|
//! Constructor with three different values
|
|
vector3d(T nx, T ny, T nz) : X(nx), Y(ny), Z(nz) {}
|
|
//! Constructor with the same value for all elements
|
|
explicit vector3d(T n) : X(n), Y(n), Z(n) {}
|
|
|
|
// operators
|
|
|
|
vector3d<T> operator-() const { return vector3d<T>(-X, -Y, -Z); }
|
|
|
|
vector3d<T> operator+(const vector3d<T>& other) const { return vector3d<T>(X + other.X, Y + other.Y, Z + other.Z); }
|
|
vector3d<T>& operator+=(const vector3d<T>& other) { X+=other.X; Y+=other.Y; Z+=other.Z; return *this; }
|
|
vector3d<T> operator+(const T val) const { return vector3d<T>(X + val, Y + val, Z + val); }
|
|
vector3d<T>& operator+=(const T val) { X+=val; Y+=val; Z+=val; return *this; }
|
|
|
|
vector3d<T> operator-(const vector3d<T>& other) const { return vector3d<T>(X - other.X, Y - other.Y, Z - other.Z); }
|
|
vector3d<T>& operator-=(const vector3d<T>& other) { X-=other.X; Y-=other.Y; Z-=other.Z; return *this; }
|
|
vector3d<T> operator-(const T val) const { return vector3d<T>(X - val, Y - val, Z - val); }
|
|
vector3d<T>& operator-=(const T val) { X-=val; Y-=val; Z-=val; return *this; }
|
|
|
|
vector3d<T> operator*(const vector3d<T>& other) const { return vector3d<T>(X * other.X, Y * other.Y, Z * other.Z); }
|
|
vector3d<T>& operator*=(const vector3d<T>& other) { X*=other.X; Y*=other.Y; Z*=other.Z; return *this; }
|
|
vector3d<T> operator*(const T v) const { return vector3d<T>(X * v, Y * v, Z * v); }
|
|
vector3d<T>& operator*=(const T v) { X*=v; Y*=v; Z*=v; return *this; }
|
|
|
|
vector3d<T> operator/(const vector3d<T>& other) const { return vector3d<T>(X / other.X, Y / other.Y, Z / other.Z); }
|
|
vector3d<T>& operator/=(const vector3d<T>& other) { X/=other.X; Y/=other.Y; Z/=other.Z; return *this; }
|
|
vector3d<T> operator/(const T v) const { return vector3d<T>(X/v, Y/v, Z/v); }
|
|
vector3d<T>& operator/=(const T v) { X/=v; Y/=v; Z/=v; return *this; }
|
|
|
|
T& operator [](u32 index)
|
|
{
|
|
_IRR_DEBUG_BREAK_IF(index>2) // access violation
|
|
|
|
return *(&X+index);
|
|
}
|
|
|
|
const T& operator [](u32 index) const
|
|
{
|
|
_IRR_DEBUG_BREAK_IF(index>2) // access violation
|
|
|
|
return *(&X+index);
|
|
}
|
|
|
|
//! sort in order X, Y, Z. Equality with rounding tolerance.
|
|
bool operator<=(const vector3d<T>&other) const
|
|
{
|
|
return (X<other.X || core::equals(X, other.X)) ||
|
|
(core::equals(X, other.X) && (Y<other.Y || core::equals(Y, other.Y))) ||
|
|
(core::equals(X, other.X) && core::equals(Y, other.Y) && (Z<other.Z || core::equals(Z, other.Z)));
|
|
}
|
|
|
|
//! sort in order X, Y, Z. Equality with rounding tolerance.
|
|
bool operator>=(const vector3d<T>&other) const
|
|
{
|
|
return (X>other.X || core::equals(X, other.X)) ||
|
|
(core::equals(X, other.X) && (Y>other.Y || core::equals(Y, other.Y))) ||
|
|
(core::equals(X, other.X) && core::equals(Y, other.Y) && (Z>other.Z || core::equals(Z, other.Z)));
|
|
}
|
|
|
|
//! sort in order X, Y, Z. Difference must be above rounding tolerance.
|
|
bool operator<(const vector3d<T>&other) const
|
|
{
|
|
return (X<other.X && !core::equals(X, other.X)) ||
|
|
(core::equals(X, other.X) && Y<other.Y && !core::equals(Y, other.Y)) ||
|
|
(core::equals(X, other.X) && core::equals(Y, other.Y) && Z<other.Z && !core::equals(Z, other.Z));
|
|
}
|
|
|
|
//! sort in order X, Y, Z. Difference must be above rounding tolerance.
|
|
bool operator>(const vector3d<T>&other) const
|
|
{
|
|
return (X>other.X && !core::equals(X, other.X)) ||
|
|
(core::equals(X, other.X) && Y>other.Y && !core::equals(Y, other.Y)) ||
|
|
(core::equals(X, other.X) && core::equals(Y, other.Y) && Z>other.Z && !core::equals(Z, other.Z));
|
|
}
|
|
|
|
//! use weak float compare
|
|
bool operator==(const vector3d<T>& other) const
|
|
{
|
|
return this->equals(other);
|
|
}
|
|
|
|
bool operator!=(const vector3d<T>& other) const
|
|
{
|
|
return !this->equals(other);
|
|
}
|
|
|
|
// functions
|
|
|
|
//! returns if this vector equals the other one, taking floating point rounding errors into account
|
|
bool equals(const vector3d<T>& other, const T tolerance = (T)ROUNDING_ERROR_f32 ) const
|
|
{
|
|
return core::equals(X, other.X, tolerance) &&
|
|
core::equals(Y, other.Y, tolerance) &&
|
|
core::equals(Z, other.Z, tolerance);
|
|
}
|
|
|
|
vector3d<T>& set(const T nx, const T ny, const T nz) {X=nx; Y=ny; Z=nz; return *this;}
|
|
vector3d<T>& set(const vector3d<T>& p) {X=p.X; Y=p.Y; Z=p.Z;return *this;}
|
|
|
|
//! Get length of the vector.
|
|
T getLength() const { return core::squareroot( X*X + Y*Y + Z*Z ); }
|
|
|
|
//! Get squared length of the vector.
|
|
/** This is useful because it is much faster than getLength().
|
|
\return Squared length of the vector. */
|
|
T getLengthSQ() const { return X*X + Y*Y + Z*Z; }
|
|
|
|
//! Get the dot product with another vector.
|
|
T dotProduct(const vector3d<T>& other) const
|
|
{
|
|
return X*other.X + Y*other.Y + Z*other.Z;
|
|
}
|
|
|
|
//! Get distance from another point.
|
|
/** Here, the vector is interpreted as point in 3 dimensional space. */
|
|
T getDistanceFrom(const vector3d<T>& other) const
|
|
{
|
|
return vector3d<T>(X - other.X, Y - other.Y, Z - other.Z).getLength();
|
|
}
|
|
|
|
//! Returns squared distance from another point.
|
|
/** Here, the vector is interpreted as point in 3 dimensional space. */
|
|
T getDistanceFromSQ(const vector3d<T>& other) const
|
|
{
|
|
return vector3d<T>(X - other.X, Y - other.Y, Z - other.Z).getLengthSQ();
|
|
}
|
|
|
|
//! Calculates the cross product with another vector.
|
|
/** \param p Vector to multiply with.
|
|
\return Crossproduct of this vector with p. */
|
|
vector3d<T> crossProduct(const vector3d<T>& p) const
|
|
{
|
|
return vector3d<T>(Y * p.Z - Z * p.Y, Z * p.X - X * p.Z, X * p.Y - Y * p.X);
|
|
}
|
|
|
|
//! Returns if this vector interpreted as a point is on a line between two other points.
|
|
/** It is assumed that the point is on the line.
|
|
\param begin Beginning vector to compare between.
|
|
\param end Ending vector to compare between.
|
|
\return True if this vector is between begin and end, false if not. */
|
|
bool isBetweenPoints(const vector3d<T>& begin, const vector3d<T>& end) const
|
|
{
|
|
const T f = (end - begin).getLengthSQ();
|
|
return getDistanceFromSQ(begin) <= f &&
|
|
getDistanceFromSQ(end) <= f;
|
|
}
|
|
|
|
//! Normalizes the vector.
|
|
/** In case of the 0 vector the result is still 0, otherwise
|
|
the length of the vector will be 1.
|
|
\return Reference to this vector after normalization. */
|
|
vector3d<T>& normalize()
|
|
{
|
|
f64 length = X*X + Y*Y + Z*Z;
|
|
if (length == 0 ) // this check isn't an optimization but prevents getting NAN in the sqrt.
|
|
return *this;
|
|
length = core::reciprocal_squareroot(length);
|
|
|
|
X = (T)(X * length);
|
|
Y = (T)(Y * length);
|
|
Z = (T)(Z * length);
|
|
return *this;
|
|
}
|
|
|
|
//! Sets the length of the vector to a new value
|
|
vector3d<T>& setLength(T newlength)
|
|
{
|
|
normalize();
|
|
return (*this *= newlength);
|
|
}
|
|
|
|
//! Inverts the vector.
|
|
vector3d<T>& invert()
|
|
{
|
|
X *= -1;
|
|
Y *= -1;
|
|
Z *= -1;
|
|
return *this;
|
|
}
|
|
|
|
//! Rotates the vector by a specified number of degrees around the Y axis and the specified center.
|
|
/** \param degrees Number of degrees to rotate around the Y axis.
|
|
\param center The center of the rotation. */
|
|
void rotateXZBy(f64 degrees, const vector3d<T>& center=vector3d<T>())
|
|
{
|
|
degrees *= DEGTORAD64;
|
|
f64 cs = cos(degrees);
|
|
f64 sn = sin(degrees);
|
|
X -= center.X;
|
|
Z -= center.Z;
|
|
set((T)(X*cs - Z*sn), Y, (T)(X*sn + Z*cs));
|
|
X += center.X;
|
|
Z += center.Z;
|
|
}
|
|
|
|
//! Rotates the vector by a specified number of degrees around the Z axis and the specified center.
|
|
/** \param degrees: Number of degrees to rotate around the Z axis.
|
|
\param center: The center of the rotation. */
|
|
void rotateXYBy(f64 degrees, const vector3d<T>& center=vector3d<T>())
|
|
{
|
|
degrees *= DEGTORAD64;
|
|
f64 cs = cos(degrees);
|
|
f64 sn = sin(degrees);
|
|
X -= center.X;
|
|
Y -= center.Y;
|
|
set((T)(X*cs - Y*sn), (T)(X*sn + Y*cs), Z);
|
|
X += center.X;
|
|
Y += center.Y;
|
|
}
|
|
|
|
//! Rotates the vector by a specified number of degrees around the X axis and the specified center.
|
|
/** \param degrees: Number of degrees to rotate around the X axis.
|
|
\param center: The center of the rotation. */
|
|
void rotateYZBy(f64 degrees, const vector3d<T>& center=vector3d<T>())
|
|
{
|
|
degrees *= DEGTORAD64;
|
|
f64 cs = cos(degrees);
|
|
f64 sn = sin(degrees);
|
|
Z -= center.Z;
|
|
Y -= center.Y;
|
|
set(X, (T)(Y*cs - Z*sn), (T)(Y*sn + Z*cs));
|
|
Z += center.Z;
|
|
Y += center.Y;
|
|
}
|
|
|
|
//! Creates an interpolated vector between this vector and another vector.
|
|
/** \param other The other vector to interpolate with.
|
|
\param d Interpolation value between 0.0f (all the other vector) and 1.0f (all this vector).
|
|
Note that this is the opposite direction of interpolation to getInterpolated_quadratic()
|
|
\return An interpolated vector. This vector is not modified. */
|
|
vector3d<T> getInterpolated(const vector3d<T>& other, f64 d) const
|
|
{
|
|
const f64 inv = 1.0 - d;
|
|
return vector3d<T>((T)(other.X*inv + X*d), (T)(other.Y*inv + Y*d), (T)(other.Z*inv + Z*d));
|
|
}
|
|
|
|
//! Creates a quadratically interpolated vector between this and two other vectors.
|
|
/** \param v2 Second vector to interpolate with.
|
|
\param v3 Third vector to interpolate with (maximum at 1.0f)
|
|
\param d Interpolation value between 0.0f (all this vector) and 1.0f (all the 3rd vector).
|
|
Note that this is the opposite direction of interpolation to getInterpolated() and interpolate()
|
|
\return An interpolated vector. This vector is not modified. */
|
|
vector3d<T> getInterpolated_quadratic(const vector3d<T>& v2, const vector3d<T>& v3, f64 d) const
|
|
{
|
|
// this*(1-d)*(1-d) + 2 * v2 * (1-d) + v3 * d * d;
|
|
const f64 inv = (T) 1.0 - d;
|
|
const f64 mul0 = inv * inv;
|
|
const f64 mul1 = (T) 2.0 * d * inv;
|
|
const f64 mul2 = d * d;
|
|
|
|
return vector3d<T> ((T)(X * mul0 + v2.X * mul1 + v3.X * mul2),
|
|
(T)(Y * mul0 + v2.Y * mul1 + v3.Y * mul2),
|
|
(T)(Z * mul0 + v2.Z * mul1 + v3.Z * mul2));
|
|
}
|
|
|
|
//! Sets this vector to the linearly interpolated vector between a and b.
|
|
/** \param a first vector to interpolate with, maximum at 1.0f
|
|
\param b second vector to interpolate with, maximum at 0.0f
|
|
\param d Interpolation value between 0.0f (all vector b) and 1.0f (all vector a)
|
|
Note that this is the opposite direction of interpolation to getInterpolated_quadratic()
|
|
*/
|
|
vector3d<T>& interpolate(const vector3d<T>& a, const vector3d<T>& b, f64 d)
|
|
{
|
|
X = (T)((f64)b.X + ( ( a.X - b.X ) * d ));
|
|
Y = (T)((f64)b.Y + ( ( a.Y - b.Y ) * d ));
|
|
Z = (T)((f64)b.Z + ( ( a.Z - b.Z ) * d ));
|
|
return *this;
|
|
}
|
|
|
|
|
|
//! Get the rotations that would make a (0,0,1) direction vector point in the same direction as this direction vector.
|
|
/** Thanks to Arras on the Irrlicht forums for this method. This utility method is very useful for
|
|
orienting scene nodes towards specific targets. For example, if this vector represents the difference
|
|
between two scene nodes, then applying the result of getHorizontalAngle() to one scene node will point
|
|
it at the other one.
|
|
Example code:
|
|
// Where target and seeker are of type ISceneNode*
|
|
const vector3df toTarget(target->getAbsolutePosition() - seeker->getAbsolutePosition());
|
|
const vector3df requiredRotation = toTarget.getHorizontalAngle();
|
|
seeker->setRotation(requiredRotation);
|
|
|
|
\return A rotation vector containing the X (pitch) and Y (raw) rotations (in degrees) that when applied to a
|
|
+Z (e.g. 0, 0, 1) direction vector would make it point in the same direction as this vector. The Z (roll) rotation
|
|
is always 0, since two Euler rotations are sufficient to point in any given direction. */
|
|
vector3d<T> getHorizontalAngle() const
|
|
{
|
|
vector3d<T> angle;
|
|
|
|
// tmp avoids some precision troubles on some compilers when working with T=s32
|
|
f64 tmp = (atan2((f64)X, (f64)Z) * RADTODEG64);
|
|
angle.Y = (T)tmp;
|
|
|
|
if (angle.Y < 0)
|
|
angle.Y += 360;
|
|
if (angle.Y >= 360)
|
|
angle.Y -= 360;
|
|
|
|
const f64 z1 = core::squareroot(X*X + Z*Z);
|
|
|
|
tmp = (atan2((f64)z1, (f64)Y) * RADTODEG64 - 90.0);
|
|
angle.X = (T)tmp;
|
|
|
|
if (angle.X < 0)
|
|
angle.X += 360;
|
|
if (angle.X >= 360)
|
|
angle.X -= 360;
|
|
|
|
return angle;
|
|
}
|
|
|
|
//! Get the spherical coordinate angles
|
|
/** This returns Euler degrees for the point represented by
|
|
this vector. The calculation assumes the pole at (0,1,0) and
|
|
returns the angles in X and Y.
|
|
*/
|
|
vector3d<T> getSphericalCoordinateAngles() const
|
|
{
|
|
vector3d<T> angle;
|
|
const f64 length = X*X + Y*Y + Z*Z;
|
|
|
|
if (length)
|
|
{
|
|
if (X!=0)
|
|
{
|
|
angle.Y = (T)(atan2((f64)Z,(f64)X) * RADTODEG64);
|
|
}
|
|
else if (Z<0)
|
|
angle.Y=180;
|
|
|
|
angle.X = (T)(acos(Y * core::reciprocal_squareroot(length)) * RADTODEG64);
|
|
}
|
|
return angle;
|
|
}
|
|
|
|
//! Builds a direction vector from (this) rotation vector.
|
|
/** This vector is assumed to be a rotation vector composed of 3 Euler angle rotations, in degrees.
|
|
The implementation performs the same calculations as using a matrix to do the rotation.
|
|
|
|
\param[in] forwards The direction representing "forwards" which will be rotated by this vector.
|
|
If you do not provide a direction, then the +Z axis (0, 0, 1) will be assumed to be forwards.
|
|
\return A direction vector calculated by rotating the forwards direction by the 3 Euler angles
|
|
(in degrees) represented by this vector. */
|
|
vector3d<T> rotationToDirection(const vector3d<T> & forwards = vector3d<T>(0, 0, 1)) const
|
|
{
|
|
const f64 cr = cos( core::DEGTORAD64 * X );
|
|
const f64 sr = sin( core::DEGTORAD64 * X );
|
|
const f64 cp = cos( core::DEGTORAD64 * Y );
|
|
const f64 sp = sin( core::DEGTORAD64 * Y );
|
|
const f64 cy = cos( core::DEGTORAD64 * Z );
|
|
const f64 sy = sin( core::DEGTORAD64 * Z );
|
|
|
|
const f64 srsp = sr*sp;
|
|
const f64 crsp = cr*sp;
|
|
|
|
const f64 pseudoMatrix[] = {
|
|
( cp*cy ), ( cp*sy ), ( -sp ),
|
|
( srsp*cy-cr*sy ), ( srsp*sy+cr*cy ), ( sr*cp ),
|
|
( crsp*cy+sr*sy ), ( crsp*sy-sr*cy ), ( cr*cp )};
|
|
|
|
return vector3d<T>(
|
|
(T)(forwards.X * pseudoMatrix[0] +
|
|
forwards.Y * pseudoMatrix[3] +
|
|
forwards.Z * pseudoMatrix[6]),
|
|
(T)(forwards.X * pseudoMatrix[1] +
|
|
forwards.Y * pseudoMatrix[4] +
|
|
forwards.Z * pseudoMatrix[7]),
|
|
(T)(forwards.X * pseudoMatrix[2] +
|
|
forwards.Y * pseudoMatrix[5] +
|
|
forwards.Z * pseudoMatrix[8]));
|
|
}
|
|
|
|
//! Fills an array of 4 values with the vector data (usually floats).
|
|
/** Useful for setting in shader constants for example. The fourth value
|
|
will always be 0. */
|
|
void getAs4Values(T* array) const
|
|
{
|
|
array[0] = X;
|
|
array[1] = Y;
|
|
array[2] = Z;
|
|
array[3] = 0;
|
|
}
|
|
|
|
//! Fills an array of 3 values with the vector data (usually floats).
|
|
/** Useful for setting in shader constants for example.*/
|
|
void getAs3Values(T* array) const
|
|
{
|
|
array[0] = X;
|
|
array[1] = Y;
|
|
array[2] = Z;
|
|
}
|
|
|
|
|
|
//! X coordinate of the vector
|
|
T X;
|
|
|
|
//! Y coordinate of the vector
|
|
T Y;
|
|
|
|
//! Z coordinate of the vector
|
|
T Z;
|
|
};
|
|
|
|
//! partial specialization for integer vectors
|
|
// Implementer note: inline keyword needed due to template specialization for s32. Otherwise put specialization into a .cpp
|
|
template <>
|
|
inline vector3d<s32> vector3d<s32>::operator /(s32 val) const {return core::vector3d<s32>(X/val,Y/val,Z/val);}
|
|
template <>
|
|
inline vector3d<s32>& vector3d<s32>::operator /=(s32 val) {X/=val;Y/=val;Z/=val; return *this;}
|
|
|
|
template <>
|
|
inline vector3d<s32> vector3d<s32>::getSphericalCoordinateAngles() const
|
|
{
|
|
vector3d<s32> angle;
|
|
const f64 length = X*X + Y*Y + Z*Z;
|
|
|
|
if (length)
|
|
{
|
|
if (X!=0)
|
|
{
|
|
angle.Y = round32((f32)(atan2((f64)Z,(f64)X) * RADTODEG64));
|
|
}
|
|
else if (Z<0)
|
|
angle.Y=180;
|
|
|
|
angle.X = round32((f32)(acos(Y * core::reciprocal_squareroot(length)) * RADTODEG64));
|
|
}
|
|
return angle;
|
|
}
|
|
|
|
//! Typedef for a f32 3d vector.
|
|
typedef vector3d<f32> vector3df;
|
|
|
|
//! Typedef for an integer 3d vector.
|
|
typedef vector3d<s32> vector3di;
|
|
|
|
//! Function multiplying a scalar and a vector component-wise.
|
|
template<class S, class T>
|
|
vector3d<T> operator*(const S scalar, const vector3d<T>& vector) { return vector*scalar; }
|
|
|
|
} // end namespace core
|
|
} // end namespace irr
|
|
|
|
namespace std
|
|
{
|
|
|
|
template<class T>
|
|
struct hash<irr::core::vector3d<T> >
|
|
{
|
|
size_t operator()(const irr::core::vector3d<T>& vec) const
|
|
{
|
|
size_t h1 = hash<T>()(vec.X);
|
|
size_t h2 = hash<T>()(vec.Y);
|
|
size_t h3 = hash<T>()(vec.Z);
|
|
return (h1 << (5 * sizeof(h1)) | h1 >> (3 * sizeof(h1))) ^ (h2 << (2 * sizeof(h2)) | h2 >> (6 * sizeof(h2))) ^ h3;
|
|
}
|
|
};
|
|
|
|
}
|
|
|
|
#endif
|
|
|