// 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_QUATERNION_H_INCLUDED #define IRR_QUATERNION_H_INCLUDED #include "irrTypes.h" #include "irrMath.h" #include "matrix4.h" #include "vector3d.h" // NOTE: You *only* need this when updating an application from Irrlicht before 1.8 to Irrlicht 1.8 or later. // Between Irrlicht 1.7 and Irrlicht 1.8 the quaternion-matrix conversions changed. // Before the fix they had mixed left- and right-handed rotations. // To test if your code was affected by the change enable IRR_TEST_BROKEN_QUATERNION_USE and try to compile your application. // This defines removes those functions so you get compile errors anywhere you use them in your code. // For every line with a compile-errors you have to change the corresponding lines like that: // - When you pass the matrix to the quaternion constructor then replace the matrix by the transposed matrix. // - For uses of getMatrix() you have to use quaternion::getMatrix_transposed instead. // #define IRR_TEST_BROKEN_QUATERNION_USE namespace irr { namespace core { //! Quaternion class for representing rotations. /** It provides cheap combinations and avoids gimbal locks. Also useful for interpolations. */ class quaternion { public: //! Default Constructor quaternion() : X(0.0f), Y(0.0f), Z(0.0f), W(1.0f) {} //! Constructor quaternion(f32 x, f32 y, f32 z, f32 w) : X(x), Y(y), Z(z), W(w) { } //! Constructor which converts Euler angles (radians) to a quaternion quaternion(f32 x, f32 y, f32 z); //! Constructor which converts Euler angles (radians) to a quaternion quaternion(const vector3df& vec); #ifndef IRR_TEST_BROKEN_QUATERNION_USE //! Constructor which converts a matrix to a quaternion quaternion(const matrix4& mat); #endif //! Equality operator bool operator==(const quaternion& other) const; //! inequality operator bool operator!=(const quaternion& other) const; #ifndef IRR_TEST_BROKEN_QUATERNION_USE //! Matrix assignment operator inline quaternion& operator=(const matrix4& other); #endif //! Add operator quaternion operator+(const quaternion& other) const; //! Multiplication operator //! Be careful, unfortunately the operator order here is opposite of that in CMatrix4::operator* quaternion operator*(const quaternion& other) const; //! Multiplication operator with scalar quaternion operator*(f32 s) const; //! Multiplication operator with scalar quaternion& operator*=(f32 s); //! Multiplication operator vector3df operator*(const vector3df& v) const; //! Multiplication operator quaternion& operator*=(const quaternion& other); //! Calculates the dot product inline f32 dotProduct(const quaternion& other) const; //! Sets new quaternion inline quaternion& set(f32 x, f32 y, f32 z, f32 w); //! Sets new quaternion based on Euler angles (radians) inline quaternion& set(f32 x, f32 y, f32 z); //! Sets new quaternion based on Euler angles (radians) inline quaternion& set(const core::vector3df& vec); //! Sets new quaternion from other quaternion inline quaternion& set(const core::quaternion& quat); //! returns if this quaternion equals the other one, taking floating point rounding errors into account inline bool equals(const quaternion& other, const f32 tolerance = ROUNDING_ERROR_f32 ) const; //! Normalizes the quaternion inline quaternion& normalize(); #ifndef IRR_TEST_BROKEN_QUATERNION_USE //! Creates a matrix from this quaternion matrix4 getMatrix() const; #endif //! Faster method to create a rotation matrix, you should normalize the quaternion before! void getMatrixFast(matrix4 &dest) const; //! Creates a matrix from this quaternion void getMatrix( matrix4 &dest, const core::vector3df &translation=core::vector3df() ) const; /*! Creates a matrix from this quaternion Rotate about a center point shortcut for core::quaternion q; q.rotationFromTo ( vin[i].Normal, forward ); q.getMatrixCenter ( lookat, center, newPos ); core::matrix4 m2; m2.setInverseTranslation ( center ); lookat *= m2; core::matrix4 m3; m2.setTranslation ( newPos ); lookat *= m3; */ void getMatrixCenter( matrix4 &dest, const core::vector3df ¢er, const core::vector3df &translation ) const; //! Faster method to create a transposed matrix, you should normalize the quaternion before! inline void getMatrixFast_transposed(matrix4 &dest) const; //! Creates a transposed matrix from this quaternion inline void getMatrix_transposed( matrix4 &dest ) const; //! Inverts this quaternion quaternion& makeInverse(); //! Set this quaternion to the linear interpolation between two quaternions /** NOTE: lerp result is *not* a normalized quaternion. In most cases you will want to use lerpN instead as most other quaternion functions expect to work with a normalized quaternion. \param q1 First quaternion to be interpolated. \param q2 Second quaternion to be interpolated. \param time Progress of interpolation. For time=0 the result is q1, for time=1 the result is q2. Otherwise interpolation between q1 and q2. Result is not normalized. */ quaternion& lerp(quaternion q1, quaternion q2, f32 time); //! Set this quaternion to the linear interpolation between two quaternions and normalize the result /** \param q1 First quaternion to be interpolated. \param q2 Second quaternion to be interpolated. \param time Progress of interpolation. For time=0 the result is q1, for time=1 the result is q2. Otherwise interpolation between q1 and q2. Result is normalized. */ quaternion& lerpN(quaternion q1, quaternion q2, f32 time); //! Set this quaternion to the result of the spherical interpolation between two quaternions /** \param q1 First quaternion to be interpolated. \param q2 Second quaternion to be interpolated. \param time Progress of interpolation. For time=0 the result is q1, for time=1 the result is q2. Otherwise interpolation between q1 and q2. \param threshold To avoid inaccuracies at the end (time=1) the interpolation switches to linear interpolation at some point. This value defines how much of the remaining interpolation will be calculated with lerp. Everything from 1-threshold up will be linear interpolation. */ quaternion& slerp(quaternion q1, quaternion q2, f32 time, f32 threshold=.05f); //! Set this quaternion to represent a rotation from angle and axis. /** Axis must be unit length. The quaternion representing the rotation is q = cos(A/2)+sin(A/2)*(x*i+y*j+z*k). \param angle Rotation Angle in radians. \param axis Rotation axis. */ quaternion& fromAngleAxis (f32 angle, const vector3df& axis); //! Fills an angle (radians) around an axis (unit vector) void toAngleAxis (f32 &angle, core::vector3df& axis) const; //! Output this quaternion to an Euler angle (radians) void toEuler(vector3df& euler) const; //! Set quaternion to identity quaternion& makeIdentity(); //! Set quaternion to represent a rotation from one vector to another. quaternion& rotationFromTo(const vector3df& from, const vector3df& to); //! Quaternion elements. f32 X; // vectorial (imaginary) part f32 Y; f32 Z; f32 W; // real part }; // Constructor which converts Euler angles to a quaternion inline quaternion::quaternion(f32 x, f32 y, f32 z) { set(x,y,z); } // Constructor which converts Euler angles to a quaternion inline quaternion::quaternion(const vector3df& vec) { set(vec.X,vec.Y,vec.Z); } #ifndef IRR_TEST_BROKEN_QUATERNION_USE // Constructor which converts a matrix to a quaternion inline quaternion::quaternion(const matrix4& mat) { (*this) = mat; } #endif // equal operator inline bool quaternion::operator==(const quaternion& other) const { return ((X == other.X) && (Y == other.Y) && (Z == other.Z) && (W == other.W)); } // inequality operator inline bool quaternion::operator!=(const quaternion& other) const { return !(*this == other); } #ifndef IRR_TEST_BROKEN_QUATERNION_USE // matrix assignment operator inline quaternion& quaternion::operator=(const matrix4& m) { const f32 diag = m[0] + m[5] + m[10] + 1; if( diag > 0.0f ) { const f32 scale = sqrtf(diag) * 2.0f; // get scale from diagonal // TODO: speed this up X = (m[6] - m[9]) / scale; Y = (m[8] - m[2]) / scale; Z = (m[1] - m[4]) / scale; W = 0.25f * scale; } else { if (m[0]>m[5] && m[0]>m[10]) { // 1st element of diag is greatest value // find scale according to 1st element, and double it const f32 scale = sqrtf(1.0f + m[0] - m[5] - m[10]) * 2.0f; // TODO: speed this up X = 0.25f * scale; Y = (m[4] + m[1]) / scale; Z = (m[2] + m[8]) / scale; W = (m[6] - m[9]) / scale; } else if (m[5]>m[10]) { // 2nd element of diag is greatest value // find scale according to 2nd element, and double it const f32 scale = sqrtf(1.0f + m[5] - m[0] - m[10]) * 2.0f; // TODO: speed this up X = (m[4] + m[1]) / scale; Y = 0.25f * scale; Z = (m[9] + m[6]) / scale; W = (m[8] - m[2]) / scale; } else { // 3rd element of diag is greatest value // find scale according to 3rd element, and double it const f32 scale = sqrtf(1.0f + m[10] - m[0] - m[5]) * 2.0f; // TODO: speed this up X = (m[8] + m[2]) / scale; Y = (m[9] + m[6]) / scale; Z = 0.25f * scale; W = (m[1] - m[4]) / scale; } } return normalize(); } #endif // multiplication operator inline quaternion quaternion::operator*(const quaternion& other) const { quaternion tmp; tmp.W = (other.W * W) - (other.X * X) - (other.Y * Y) - (other.Z * Z); tmp.X = (other.W * X) + (other.X * W) + (other.Y * Z) - (other.Z * Y); tmp.Y = (other.W * Y) + (other.Y * W) + (other.Z * X) - (other.X * Z); tmp.Z = (other.W * Z) + (other.Z * W) + (other.X * Y) - (other.Y * X); return tmp; } // multiplication operator inline quaternion quaternion::operator*(f32 s) const { return quaternion(s*X, s*Y, s*Z, s*W); } // multiplication operator inline quaternion& quaternion::operator*=(f32 s) { X*=s; Y*=s; Z*=s; W*=s; return *this; } // multiplication operator inline quaternion& quaternion::operator*=(const quaternion& other) { return (*this = other * (*this)); } // add operator inline quaternion quaternion::operator+(const quaternion& b) const { return quaternion(X+b.X, Y+b.Y, Z+b.Z, W+b.W); } #ifndef IRR_TEST_BROKEN_QUATERNION_USE // Creates a matrix from this quaternion inline matrix4 quaternion::getMatrix() const { core::matrix4 m; getMatrix(m); return m; } #endif //! Faster method to create a rotation matrix, you should normalize the quaternion before! inline void quaternion::getMatrixFast( matrix4 &dest) const { // TODO: // gpu quaternion skinning => fast Bones transform chain O_O YEAH! // http://www.mrelusive.com/publications/papers/SIMD-From-Quaternion-to-Matrix-and-Back.pdf dest[0] = 1.0f - 2.0f*Y*Y - 2.0f*Z*Z; dest[1] = 2.0f*X*Y + 2.0f*Z*W; dest[2] = 2.0f*X*Z - 2.0f*Y*W; dest[3] = 0.0f; dest[4] = 2.0f*X*Y - 2.0f*Z*W; dest[5] = 1.0f - 2.0f*X*X - 2.0f*Z*Z; dest[6] = 2.0f*Z*Y + 2.0f*X*W; dest[7] = 0.0f; dest[8] = 2.0f*X*Z + 2.0f*Y*W; dest[9] = 2.0f*Z*Y - 2.0f*X*W; dest[10] = 1.0f - 2.0f*X*X - 2.0f*Y*Y; dest[11] = 0.0f; dest[12] = 0.f; dest[13] = 0.f; dest[14] = 0.f; dest[15] = 1.f; dest.setDefinitelyIdentityMatrix(false); } /*! Creates a matrix from this quaternion */ inline void quaternion::getMatrix(matrix4 &dest, const core::vector3df ¢er) const { // ok creating a copy may be slower, but at least avoid internal // state chance (also because otherwise we cannot keep this method "const"). quaternion q( *this); q.normalize(); q.getMatrixFast(dest); dest[12] = center.X; dest[13] = center.Y; dest[14] = center.Z; } /*! Creates a matrix from this quaternion Rotate about a center point shortcut for core::quaternion q; q.rotationFromTo(vin[i].Normal, forward); q.getMatrix(lookat, center); core::matrix4 m2; m2.setInverseTranslation(center); lookat *= m2; */ inline void quaternion::getMatrixCenter(matrix4 &dest, const core::vector3df ¢er, const core::vector3df &translation) const { quaternion q(*this); q.normalize(); q.getMatrixFast(dest); dest.setRotationCenter ( center, translation ); } //! Faster method to create a transposed matrix, you should normalize the quaternion before! inline void quaternion::getMatrixFast_transposed(matrix4 &dest) const { dest[0] = 1.0f - 2.0f*Y*Y - 2.0f*Z*Z; dest[4] = 2.0f*X*Y + 2.0f*Z*W; dest[8] = 2.0f*X*Z - 2.0f*Y*W; dest[12] = 0.0f; dest[1] = 2.0f*X*Y - 2.0f*Z*W; dest[5] = 1.0f - 2.0f*X*X - 2.0f*Z*Z; dest[9] = 2.0f*Z*Y + 2.0f*X*W; dest[13] = 0.0f; dest[2] = 2.0f*X*Z + 2.0f*Y*W; dest[6] = 2.0f*Z*Y - 2.0f*X*W; dest[10] = 1.0f - 2.0f*X*X - 2.0f*Y*Y; dest[14] = 0.0f; dest[3] = 0.f; dest[7] = 0.f; dest[11] = 0.f; dest[15] = 1.f; dest.setDefinitelyIdentityMatrix(false); } // Creates a matrix from this quaternion inline void quaternion::getMatrix_transposed(matrix4 &dest) const { quaternion q(*this); q.normalize(); q.getMatrixFast_transposed(dest); } // Inverts this quaternion inline quaternion& quaternion::makeInverse() { X = -X; Y = -Y; Z = -Z; return *this; } // sets new quaternion inline quaternion& quaternion::set(f32 x, f32 y, f32 z, f32 w) { X = x; Y = y; Z = z; W = w; return *this; } // sets new quaternion based on Euler angles inline quaternion& quaternion::set(f32 x, f32 y, f32 z) { f64 angle; angle = x * 0.5; const f64 sr = sin(angle); const f64 cr = cos(angle); angle = y * 0.5; const f64 sp = sin(angle); const f64 cp = cos(angle); angle = z * 0.5; const f64 sy = sin(angle); const f64 cy = cos(angle); const f64 cpcy = cp * cy; const f64 spcy = sp * cy; const f64 cpsy = cp * sy; const f64 spsy = sp * sy; X = (f32)(sr * cpcy - cr * spsy); Y = (f32)(cr * spcy + sr * cpsy); Z = (f32)(cr * cpsy - sr * spcy); W = (f32)(cr * cpcy + sr * spsy); return normalize(); } // sets new quaternion based on Euler angles inline quaternion& quaternion::set(const core::vector3df& vec) { return set( vec.X, vec.Y, vec.Z); } // sets new quaternion based on other quaternion inline quaternion& quaternion::set(const core::quaternion& quat) { return (*this=quat); } //! returns if this quaternion equals the other one, taking floating point rounding errors into account inline bool quaternion::equals(const quaternion& other, const f32 tolerance) const { return core::equals( X, other.X, tolerance) && core::equals( Y, other.Y, tolerance) && core::equals( Z, other.Z, tolerance) && core::equals( W, other.W, tolerance); } // normalizes the quaternion inline quaternion& quaternion::normalize() { // removed conditional branch since it may slow down and anyway the condition was // false even after normalization in some cases. return (*this *= (f32)reciprocal_squareroot ( (f64)(X*X + Y*Y + Z*Z + W*W) )); } // Set this quaternion to the result of the linear interpolation between two quaternions inline quaternion& quaternion::lerp( quaternion q1, quaternion q2, f32 time) { const f32 scale = 1.0f - time; return (*this = (q1*scale) + (q2*time)); } // Set this quaternion to the result of the linear interpolation between two quaternions and normalize the result inline quaternion& quaternion::lerpN( quaternion q1, quaternion q2, f32 time) { const f32 scale = 1.0f - time; return (*this = ((q1*scale) + (q2*time)).normalize() ); } // set this quaternion to the result of the interpolation between two quaternions inline quaternion& quaternion::slerp( quaternion q1, quaternion q2, f32 time, f32 threshold) { f32 angle = q1.dotProduct(q2); // make sure we use the short rotation if (angle < 0.0f) { q1 *= -1.0f; angle *= -1.0f; } if (angle <= (1-threshold)) // spherical interpolation { const f32 theta = acosf(angle); const f32 invsintheta = reciprocal(sinf(theta)); const f32 scale = sinf(theta * (1.0f-time)) * invsintheta; const f32 invscale = sinf(theta * time) * invsintheta; return (*this = (q1*scale) + (q2*invscale)); } else // linear interpolation return lerpN(q1,q2,time); } // calculates the dot product inline f32 quaternion::dotProduct(const quaternion& q2) const { return (X * q2.X) + (Y * q2.Y) + (Z * q2.Z) + (W * q2.W); } //! axis must be unit length, angle in radians inline quaternion& quaternion::fromAngleAxis(f32 angle, const vector3df& axis) { const f32 fHalfAngle = 0.5f*angle; const f32 fSin = sinf(fHalfAngle); W = cosf(fHalfAngle); X = fSin*axis.X; Y = fSin*axis.Y; Z = fSin*axis.Z; return *this; } inline void quaternion::toAngleAxis(f32 &angle, core::vector3df &axis) const { const f32 scale = sqrtf(X*X + Y*Y + Z*Z); if (core::iszero(scale) || W > 1.0f || W < -1.0f) { angle = 0.0f; axis.X = 0.0f; axis.Y = 1.0f; axis.Z = 0.0f; } else { const f32 invscale = reciprocal(scale); angle = 2.0f * acosf(W); axis.X = X * invscale; axis.Y = Y * invscale; axis.Z = Z * invscale; } } inline void quaternion::toEuler(vector3df& euler) const { const f64 sqw = W*W; const f64 sqx = X*X; const f64 sqy = Y*Y; const f64 sqz = Z*Z; const f64 test = 2.0 * (Y*W - X*Z); if (core::equals(test, 1.0, 0.000001)) { // heading = rotation about z-axis euler.Z = (f32) (-2.0*atan2(X, W)); // bank = rotation about x-axis euler.X = 0; // attitude = rotation about y-axis euler.Y = (f32) (core::PI64/2.0); } else if (core::equals(test, -1.0, 0.000001)) { // heading = rotation about z-axis euler.Z = (f32) (2.0*atan2(X, W)); // bank = rotation about x-axis euler.X = 0; // attitude = rotation about y-axis euler.Y = (f32) (core::PI64/-2.0); } else { // heading = rotation about z-axis euler.Z = (f32) atan2(2.0 * (X*Y +Z*W),(sqx - sqy - sqz + sqw)); // bank = rotation about x-axis euler.X = (f32) atan2(2.0 * (Y*Z +X*W),(-sqx - sqy + sqz + sqw)); // attitude = rotation about y-axis euler.Y = (f32) asin( clamp(test, -1.0, 1.0) ); } } inline vector3df quaternion::operator* (const vector3df& v) const { // nVidia SDK implementation vector3df uv, uuv; const vector3df qvec(X, Y, Z); uv = qvec.crossProduct(v); uuv = qvec.crossProduct(uv); uv *= (2.0f * W); uuv *= 2.0f; return v + uv + uuv; } // set quaternion to identity inline core::quaternion& quaternion::makeIdentity() { W = 1.f; X = 0.f; Y = 0.f; Z = 0.f; return *this; } inline core::quaternion& quaternion::rotationFromTo(const vector3df& from, const vector3df& to) { // Based on Stan Melax's article in Game Programming Gems // Optimized by Robert Eisele: https://raw.org/proof/quaternion-from-two-vectors // Copy, since cannot modify local vector3df v0 = from; vector3df v1 = to; v0.normalize(); v1.normalize(); const f32 d = v0.dotProduct(v1); if (d >= 1.0f) // If dot == 1, vectors are the same { return makeIdentity(); } else if (d <= -1.0f) // exactly opposite { core::vector3df axis(1.0f, 0.f, 0.f); axis = axis.crossProduct(v0); if (axis.getLength()==0) { axis.set(0.f,1.f,0.f); axis = axis.crossProduct(v0); } // same as fromAngleAxis(core::PI, axis).normalize(); return set(axis.X, axis.Y, axis.Z, 0).normalize(); } const vector3df c = v0.crossProduct(v1); return set(c.X, c.Y, c.Z, 1 + d).normalize(); } } // end namespace core } // end namespace irr #endif