bluecore/engine/Math/Matrix.h
2008-01-16 11:45:17 +00:00

867 lines
20 KiB
C++

#ifndef BLUECORE_MATRIX_H
#define BLUECORE_MATRIX_H
// system includes
#include <algorithm>
#include <cmath>
// local includes
#include "Scalar.h"
#include "Vector.h"
#include "Quaternion.h"
namespace BlueCore
{
template <class T>
class Matrix3x3Template
{
public:
T m[9];
/**
* default constructor
*/
inline Matrix3x3Template()
{
identity();
}
/**
* constructor from array
*/
template <class S>
inline Matrix3x3Template( const S a[9] )
{
m[0] = static_cast<T>(a[0]);
m[1] = static_cast<T>(a[1]);
m[2] = static_cast<T>(a[2]);
m[3] = static_cast<T>(a[3]);
m[4] = static_cast<T>(a[4]);
m[5] = static_cast<T>(a[5]);
m[6] = static_cast<T>(a[6]);
m[7] = static_cast<T>(a[7]);
m[8] = static_cast<T>(a[8]);
}
/**
* copy constructor
*/
template <class S>
inline Matrix3x3Template( const Matrix3x3Template<S> &a )
{
m[0] = static_cast<T>(a.m[0]);
m[1] = static_cast<T>(a.m[1]);
m[2] = static_cast<T>(a.m[2]);
m[3] = static_cast<T>(a.m[3]);
m[4] = static_cast<T>(a.m[4]);
m[5] = static_cast<T>(a.m[5]);
m[6] = static_cast<T>(a.m[6]);
m[7] = static_cast<T>(a.m[7]);
m[8] = static_cast<T>(a.m[8]);
}
/**
* constructor from heading and up vectors
*/
inline Matrix3x3Template(
const Vector3Template<T> &h,
const Vector3Template<T> &u )
{
Vector3Template<T> a, b, c;
c = h.unit();
a = h.crossProduct( u ).unit();
b = c.crossProduct( a ).unit();
m[0] = a.x;
m[1] = b.x;
m[2] = c.x;
m[3] = a.y;
m[4] = b.y;
m[5] = c.y;
m[6] = a.z;
m[7] = b.z;
m[8] = c.z;
}
/**
* contructor from euler angles
*/
inline Matrix3x3Template( T a, T e, T t )
{
T ch = cos( a );
T sh = sin( a );
T ca = cos( e );
T sa = sin( e );
T cb = cos( t );
T sb = sin( t );
m[0] = ch * ca;
m[1] = sh*sb - ch*sa*cb;
m[2] = ch*sa*sb + sh*cb;
m[3] = sa;
m[4] = ca*cb;
m[5] = -ca*sb;
m[6] = -sh*ca;
m[7] = sh*sa*cb + ch*sb;
m[8] = -sh*sa*sb + ch*cb;
}
/**
* contructor from quaternion
*/
template <class S>
inline Matrix3x3Template( const QuaternionTemplate<S> &q )
{
T xx = static_cast<T>(q.x*q.x),
xy = static_cast<T>(q.x*q.y),
xz = static_cast<T>(q.x*q.z),
xw = static_cast<T>(q.x*q.w),
yy = static_cast<T>(q.y*q.y),
yz = static_cast<T>(q.y*q.z),
yw = static_cast<T>(q.y*q.w),
zz = static_cast<T>(q.z*q.z),
zw = static_cast<T>(q.z*q.w);
m[0] = 1 - yy - yy - zz - zz;
m[1] = xy + xy + zw + zw;
m[2] = xz + xz - yw - yw;
m[3] = xy + xy - zw - zw;
m[4] = 1 - xx - xx - zz - zz;
m[5] = yz + yz + xw + xw;
m[6] = xz + xz + yw + yw;
m[7] = yz + yz - xw - xw;
m[8] = 1 - xx - xx - yy - yy;
}
/**
* set matrix to identity matrix
*/
inline void identity()
{
m[0] = static_cast<T>(1);
m[1] = static_cast<T>(0);
m[2] = static_cast<T>(0);
m[3] = static_cast<T>(0);
m[4] = static_cast<T>(1);
m[5] = static_cast<T>(0);
m[6] = static_cast<T>(0);
m[7] = static_cast<T>(0);
m[8] = static_cast<T>(1);
}
/**
* get transposed matrix
*/
inline Matrix3x3Template<T> transposed() const
{
Matrix3x3Template<T> a;
a.m[0] = m[0]; a.m[3] = m[1]; a.m[6] = m[2];
a.m[1] = m[3]; a.m[4] = m[4]; a.m[7] = m[5];
a.m[2] = m[6]; a.m[5] = m[7]; a.m[8] = m[8];
return a;
}
/**
* transpose
*/
inline void transpose()
{
swap( m[3], m[1] );
swap( m[6], m[2] );
swap( m[7], m[5] );
}
/**
* determinate
*/
inline T determinant() const
{
return ( m[0]*m[4]*m[8] + m[1]*m[5]*m[6] + m[2]*m[3]*m[7] )
- ( m[6]*m[4]*m[2] + m[7]*m[5]*m[0] + m[8]*m[3]*m[1] );
}
inline Matrix3x3Template<T> inverted()
{
T det = determinant();
T one_over_det = 1.0f / det;
Matrix3x3Template<T> result;
result.m[0] = +(m[4] * m[8] - m[5] * m[7]) * one_over_det;
result.m[1] = -(m[1] * m[8] - m[2] * m[7]) * one_over_det;
result.m[2] = +(m[1] * m[5] - m[2] * m[4]) * one_over_det;
result.m[3] = -(m[3] * m[8] - m[5] * m[6]) * one_over_det;
result.m[4] = +(m[0] * m[8] - m[2] * m[6]) * one_over_det;
result.m[5] = -(m[0] * m[5] - m[2] * m[3]) * one_over_det;
result.m[6] = +(m[3] * m[7] - m[4] * m[6]) * one_over_det;
result.m[7] = -(m[0] * m[7] - m[1] * m[6]) * one_over_det;
result.m[8] = +(m[0] * m[4] - m[1] * m[3]) * one_over_det;
return result;
}
/**
* add/assign operator
*/
template <class S>
inline Matrix3x3Template<T> &operator += (
const Matrix3x3Template<S> &a )
{
m[0] += static_cast<T>(a.m[0]);
m[1] += static_cast<T>(a.m[1]);
m[2] += static_cast<T>(a.m[2]);
m[3] += static_cast<T>(a.m[3]);
m[4] += static_cast<T>(a.m[4]);
m[5] += static_cast<T>(a.m[5]);
m[6] += static_cast<T>(a.m[6]);
m[7] += static_cast<T>(a.m[7]);
m[8] += static_cast<T>(a.m[8]);
return *this;
}
/**
* matrix multiplication
*/
template <class S>
inline Matrix3x3Template<T> operator * ( const Matrix3x3Template<S> &a )
{
Matrix3x3Template<T> result;
result.m[0] = m[0] * static_cast<T>(a.m[0])
+ m[3] * static_cast<T>(a.m[1])
+ m[6] * static_cast<T>(a.m[2]);
result.m[1] = m[1] * static_cast<T>(a.m[0])
+ m[4] * static_cast<T>(a.m[1])
+ m[7] * static_cast<T>(a.m[2]);
result.m[2] = m[2] * static_cast<T>(a.m[0])
+ m[5] * static_cast<T>(a.m[1])
+ m[8] * static_cast<T>(a.m[2]);
result.m[3] = m[0] * static_cast<T>(a.m[3])
+ m[3] * static_cast<T>(a.m[4])
+ m[6] * static_cast<T>(a.m[5]);
result.m[4] = m[1] * static_cast<T>(a.m[3])
+ m[4] * static_cast<T>(a.m[4])
+ m[7] * static_cast<T>(a.m[5]);
result.m[5] = m[2] * static_cast<T>(a.m[3])
+ m[5] * static_cast<T>(a.m[4])
+ m[8] * static_cast<T>(a.m[5]);
result.m[6] = m[0] * static_cast<T>(a.m[6])
+ m[3] * static_cast<T>(a.m[7])
+ m[6] * static_cast<T>(a.m[8]);
result.m[7] = m[1] * static_cast<T>(a.m[6])
+ m[4] * static_cast<T>(a.m[7])
+ m[7] * static_cast<T>(a.m[8]);
result.m[8] = m[2] * static_cast<T>(a.m[6])
+ m[5] * static_cast<T>(a.m[7])
+ m[8] * static_cast<T>(a.m[8]);
return result;
}
/**
* matrix vector multiplication
*/
template <class S>
Vector3Template<T> operator * ( const Vector3Template<S>& a )
{
return Vector3Template<T>(
m[0] * static_cast<T>(a.x)
+ m[3] * static_cast<T>(a.y)
+ m[6] * static_cast<T>(a.z),
m[1] * static_cast<T>(a.x)
+ m[4] * static_cast<T>(a.y)
+ m[7] * static_cast<T>(a.z),
m[2] * static_cast<T>(a.x)
+ m[5] * static_cast<T>(a.y)
+ m[8] * static_cast<T>(a.z) );
}
/**
* matrix scalar multiplication
*/
template <class S>
Matrix3x3Template<T> operator * ( const S a )
{
Matrix3x3Template<T> result;
result.m[0] = m[0] * static_cast<T>(a);
result.m[1] = m[1] * static_cast<T>(a);
result.m[2] = m[2] * static_cast<T>(a);
result.m[3] = m[3] * static_cast<T>(a);
result.m[4] = m[4] * static_cast<T>(a);
result.m[5] = m[5] * static_cast<T>(a);
result.m[6] = m[6] * static_cast<T>(a);
result.m[7] = m[7] * static_cast<T>(a);
result.m[8] = m[8] * static_cast<T>(a);
return result;
}
friend std::ostream &operator << ( std::ostream& os, Matrix3x3Template<T> m )
{
os << "( " << m.m[0] << ", " << m.m[1] << ", " << m.m[2] << " )" << std::endl;
os << "( " << m.m[3] << ", " << m.m[4] << ", " << m.m[5] << " )" << std::endl;
os << "( " << m.m[6] << ", " << m.m[7] << ", " << m.m[8] << " )" << std::endl;
return os;
}
};
typedef Matrix3x3Template<float> Matrix3x3Float;
typedef Matrix3x3Template<double> Matrix3x3Double;
typedef Matrix3x3Template<Scalar> Matrix3x3;
template <class T>
class Matrix4x4Template
{
public:
T m[16];
/**
* default constructor
*/
inline Matrix4x4Template()
{
identity();
}
inline void identity()
{
m[0] = static_cast<T>(1);
m[1] = static_cast<T>(0);
m[2] = static_cast<T>(0);
m[3] = static_cast<T>(0);
m[4] = static_cast<T>(0);
m[5] = static_cast<T>(1);
m[6] = static_cast<T>(0);
m[7] = static_cast<T>(0);
m[8] = static_cast<T>(0);
m[9] = static_cast<T>(0);
m[10] = static_cast<T>(1);
m[11] = static_cast<T>(0);
m[12] = static_cast<T>(0);
m[13] = static_cast<T>(0);
m[14] = static_cast<T>(0);
m[15] = static_cast<T>(1);
}
/**
* constructor from array
*/
inline Matrix4x4Template( const T a[16] )
{
m[0] = a[0]; m[1] = a[1]; m[2] = a[2]; m[3] = a[3];
m[4] = a[4]; m[5] = a[5]; m[6] = a[6]; m[7] = a[7];
m[8] = a[8]; m[9] = a[9]; m[10] = a[10]; m[11] = a[11];
m[12] = a[12]; m[13] = a[13]; m[14] = a[14]; m[15] = a[15];
}
/**
* constructor from 3x3 matrix, add. element are set to zero
*/
inline Matrix4x4Template( const Matrix3x3Template<T> &a )
{
m[0] = a[0];
m[1] = a[1];
m[2] = a[2];
m[3] = static_cast<T>(0);
m[4] = a[3];
m[5] = a[4];
m[6] = a[5];
m[7] = static_cast<T>(0);
m[8] = a[6];
m[9] = a[7];
m[10] = a[8];
m[11] = static_cast<T>(0);
m[12] = static_cast<T>(0);
m[13] = static_cast<T>(0);
m[14] = static_cast<T>(0);
m[15] = static_cast<T>(1);
}
/**
* constructor from 3x3 rotation matrix and translation vector
*/
inline Matrix4x4Template(
const Matrix3x3Template<T> &a,
const Vector3Template<T> &b )
{
m[0] = a[0];
m[1] = a[1];
m[2] = a[2];
m[3] = static_cast<T>(0);
m[4] = a[3];
m[5] = a[4];
m[6] = a[5];
m[7] = static_cast<T>(0);
m[8] = a[6];
m[9] = a[7];
m[10] = a[8];
m[11] = static_cast<T>(0);
m[12] = b.x;
m[13] = b.y;
m[14] = b.z;
m[15] = static_cast<T>(1);
}
/**
* copy constructor
*/
inline Matrix4x4Template( const Matrix4x4Template<T> &a )
{
m[0] = a.m[0];
m[1] = a.m[1];
m[2] = a.m[2];
m[3] = a.m[3];
m[4] = a.m[4];
m[5] = a.m[5];
m[6] = a.m[6];
m[7] = a.m[7];
m[8] = a.m[8];
m[9] = a.m[9];
m[10] = a.m[10];
m[11] = a.m[11];
m[12] = a.m[12];
m[13] = a.m[13];
m[14] = a.m[14];
m[15] = a.m[15];
}
/**
* constructor from quaternion
*/
inline Matrix4x4Template( const QuaternionTemplate<T> &q )
{
m[0] = 1 - 2*q.y*q.y - 2*q.z*q.z;
m[1] = 2*q.x*q.y - 2*q.z*q.w;
m[2] = 2*q.x*q.z + 2*q.y*q.w;
m[3] = static_cast<T>(0);
m[4] = 2*q.x*q.y + 2*q.z*q.w;
m[5] = 1 - 2*q.x*q.x - 2*q.z*q.z;
m[6] = 2*q.y*q.z - 2*q.x*q.w;
m[7] = static_cast<T>(0);
m[8] = 2*q.x*q.z - 2*q.y*q.w;
m[9] = 2*q.y*q.z + 2*q.x*q.w;
m[10] = 1 - 2*q.x*q.x - 2*q.y*q.y;
m[11] = static_cast<T>(0);
m[12] = static_cast<T>(0);
m[13] = static_cast<T>(0);
m[14] = static_cast<T>(0);
m[15] = static_cast<T>(1);
}
/**
* constructor from quaternion and translation vector
*/
inline Matrix4x4Template(
const QuaternionTemplate<T> &q,
const Vector3Template<T> &a )
{
m[0] = 1 - 2*q.y*q.y - 2*q.z*q.z;
m[1] = 2*q.x*q.y - 2*q.z*q.w;
m[2] = 2*q.x*q.z + 2*q.y*q.w;
m[3] = static_cast<T>(0);
m[4] = 2*q.x*q.y + 2*q.z*q.w;
m[5] = 1 - 2*q.x*q.x - 2*q.z*q.z;
m[6] = 2*q.y*q.z - 2*q.x*q.w;
m[7] = static_cast<T>(0);
m[8] = 2*q.x*q.z - 2*q.y*q.w;
m[9] = 2*q.y*q.z + 2*q.x*q.w;
m[10] = 1 - 2*q.x*q.x - 2*q.y*q.y;
m[11] = static_cast<T>(0);
m[12] = a.x;
m[13] = a.y;
m[14] = a.z;
m[15] = static_cast<T>(1);
}
/**
* constructor from heading, up and translation vectors
*/
inline Matrix4x4Template(
const Vector3Template<T> &h,
const Vector3Template<T> &u,
const Vector3Template<T> &t )
{
Vector3Template<T> a, b, c;
c = h.unit();
a = h.crossProduct( u ).unit();
b = c.crossProduct( a ).unit();
m[0] = a.x;
m[1] = b.x;
m[2] = c.x;
m[3] = static_cast<T>(0);
m[4] = a.y;
m[5] = b.y;
m[6] = c.y;
m[7] = static_cast<T>(0);
m[8] = a.z;
m[9] = b.z;
m[10] = c.z;
m[11] = static_cast<T>(0);
m[12] = t.x;
m[13] = t.y;
m[14] = t.z;
m[15] = static_cast<T>(1);
}
/**
* contructor from euler angles
*/
inline Matrix4x4Template(
T a,
T e,
T t,
const Vector3Template<T> &tr )
{
T ch = cos( a );
T sh = sin( a );
T ca = cos( e );
T sa = sin( e );
T cb = cos( t );
T sb = sin( t );
m[0] = ch * ca;
m[1] = sh*sb - ch*sa*cb;
m[2] = ch*sa*sb + sh*cb;
m[3] = static_cast<T>(0);
m[4] = sa;
m[5] = ca*cb;
m[6] = -ca*sb;
m[7] = static_cast<T>(0);
m[8] = -sh*ca;
m[9] = sh*sa*cb + ch*sb;
m[10] = -sh*sa*sb + ch*cb;
m[11] = static_cast<T>(0);
m[12] = tr.x;
m[13] = tr.y;
m[14] = tr.z;
m[15] = static_cast<T>(1);
}
/**
* transpose this matrix
*/
inline void transpose()
{
std::swap( m[4], m[1] );
std::swap( m[8], m[2] );
std::swap( m[12], m[3] );
std::swap( m[9], m[6] );
std::swap( m[13], m[7] );
std::swap( m[14], m[11] );
}
/**
* get transposed matrix
*/
inline Matrix4x4Template<T> transposed()
{
Matrix4x4Template<T> a;
a.m[0] = m[0]; a.m[4] = m[1]; a.m[8] = m[2]; a.m[12] = m[3];
a.m[1] = m[4]; a.m[5] = m[5]; a.m[9] = m[6]; a.m[13] = m[7];
a.m[2] = m[8]; a.m[6] = m[9]; a.m[10] = m[10]; a.m[14] = m[11];
a.m[3] = m[12]; a.m[7] = m[13]; a.m[11] = m[14]; a.m[15] = m[15];
return a;
}
/**
* matrix multiplication
*/
template <class S>
inline Matrix4x4Template<T> operator * ( const Matrix4x4Template<S> &a )
{
Matrix4x4Template<T> result;
result.m[0] = m[0] * static_cast<T>(a.m[0])
+ m[4] * static_cast<T>(a.m[1])
+ m[8] * static_cast<T>(a.m[2])
+ m[12] * static_cast<T>(a.m[3]);
result.m[1] = m[1] * static_cast<T>(a.m[0])
+ m[5] * static_cast<T>(a.m[1])
+ m[9] * static_cast<T>(a.m[2])
+ m[13] * static_cast<T>(a.m[3]);
result.m[2] = m[2] * static_cast<T>(a.m[0])
+ m[6] * static_cast<T>(a.m[1])
+ m[10] * static_cast<T>(a.m[2])
+ m[14] * static_cast<T>(a.m[3]);
result.m[3] = m[3] * static_cast<T>(a.m[0])
+ m[7] * static_cast<T>(a.m[1])
+ m[11] * static_cast<T>(a.m[2])
+ m[15] * static_cast<T>(a.m[3]);
result.m[4] = m[0] * static_cast<T>(a.m[4])
+ m[4] * static_cast<T>(a.m[5])
+ m[8] * static_cast<T>(a.m[6])
+ m[12] * static_cast<T>(a.m[7]);
result.m[5] = m[1] * static_cast<T>(a.m[4])
+ m[5] * static_cast<T>(a.m[5])
+ m[9] * static_cast<T>(a.m[6])
+ m[13] * static_cast<T>(a.m[7]);
result.m[6] = m[2] * static_cast<T>(a.m[4])
+ m[6] * static_cast<T>(a.m[5])
+ m[10] * static_cast<T>(a.m[6])
+ m[14] * static_cast<T>(a.m[7]);
result.m[7] = m[3] * static_cast<T>(a.m[4])
+ m[7] * static_cast<T>(a.m[5])
+ m[11] * static_cast<T>(a.m[6])
+ m[15] * static_cast<T>(a.m[7]);
result.m[8] = m[0] * static_cast<T>(a.m[8])
+ m[4] * static_cast<T>(a.m[9])
+ m[8] * static_cast<T>(a.m[10])
+ m[12] * static_cast<T>(a.m[11]);
result.m[9] = m[1] * static_cast<T>(a.m[8])
+ m[5] * static_cast<T>(a.m[9])
+ m[9] * static_cast<T>(a.m[10])
+ m[13] * static_cast<T>(a.m[11]);
result.m[10] = m[2] * static_cast<T>(a.m[8])
+ m[6] * static_cast<T>(a.m[9])
+ m[10] * static_cast<T>(a.m[10])
+ m[14] * static_cast<T>(a.m[11]);
result.m[11] = m[3] * static_cast<T>(a.m[8])
+ m[7] * static_cast<T>(a.m[9])
+ m[11] * static_cast<T>(a.m[10])
+ m[15] * static_cast<T>(a.m[11]);
result.m[12] = m[0] * static_cast<T>(a.m[12])
+ m[4] * static_cast<T>(a.m[13])
+ m[8] * static_cast<T>(a.m[14])
+ m[12] * static_cast<T>(a.m[15]);
result.m[13] = m[1] * static_cast<T>(a.m[12])
+ m[5] * static_cast<T>(a.m[13])
+ m[9] * static_cast<T>(a.m[14])
+ m[13] * static_cast<T>(a.m[15]);
result.m[14] = m[2] * static_cast<T>(a.m[12])
+ m[6] * static_cast<T>(a.m[13])
+ m[10] * static_cast<T>(a.m[14])
+ m[14] * static_cast<T>(a.m[15]);
result.m[15] = m[3] * static_cast<T>(a.m[12])
+ m[7] * static_cast<T>(a.m[13])
+ m[11] * static_cast<T>(a.m[14])
+ m[15] * static_cast<T>(a.m[15]);
return result;
}
/**
* matrix multiplication
*/
/*
template <class S>
inline Matrix4x4Template<T> operator *= ( const Matrix4x4Template<S> &a )
{
Matrix4x4Template<T> result;
result.m[0] = m[0] * static_cast<T>(a.m[0])
+ m[4] * static_cast<T>(a.m[1])
+ m[8] * static_cast<T>(a.m[2]);
+ m[12] * static_cast<T>(a.m[3]);
result.m[1] = m[1] * static_cast<T>(a.m[0])
+ m[5] * static_cast<T>(a.m[1])
+ m[9] * static_cast<T>(a.m[2]);
+ m[13] * static_cast<T>(a.m[3]);
result.m[2] = m[2] * static_cast<T>(a.m[0])
+ m[6] * static_cast<T>(a.m[1])
+ m[10] * static_cast<T>(a.m[2]);
+ m[14] * static_cast<T>(a.m[3]);
result.m[3] = m[3] * static_cast<T>(a.m[0])
+ m[7] * static_cast<T>(a.m[1])
+ m[11] * static_cast<T>(a.m[2]);
+ m[15] * static_cast<T>(a.m[3]);
result.m[4] = m[0] * static_cast<T>(a.m[4])
+ m[4] * static_cast<T>(a.m[5])
+ m[8] * static_cast<T>(a.m[6]);
+ m[12] * static_cast<T>(a.m[7]);
result.m[5] = m[1] * static_cast<T>(a.m[4])
+ m[5] * static_cast<T>(a.m[5])
+ m[9] * static_cast<T>(a.m[6]);
+ m[13] * static_cast<T>(a.m[7]);
result.m[6] = m[2] * static_cast<T>(a.m[4])
+ m[6] * static_cast<T>(a.m[5])
+ m[10] * static_cast<T>(a.m[6]);
+ m[14] * static_cast<T>(a.m[7]);
result.m[7] = m[3] * static_cast<T>(a.m[4])
+ m[7] * static_cast<T>(a.m[5])
+ m[11] * static_cast<T>(a.m[6]);
+ m[15] * static_cast<T>(a.m[7]);
result.m[8] = m[0] * static_cast<T>(a.m[8])
+ m[4] * static_cast<T>(a.m[9])
+ m[8] * static_cast<T>(a.m[10]);
+ m[12] * static_cast<T>(a.m[11]);
result.m[9] = m[1] * static_cast<T>(a.m[8])
+ m[5] * static_cast<T>(a.m[9])
+ m[9] * static_cast<T>(a.m[10]);
+ m[13] * static_cast<T>(a.m[11]);
result.m[10] = m[2] * static_cast<T>(a.m[8])
+ m[6] * static_cast<T>(a.m[9])
+ m[10] * static_cast<T>(a.m[10]);
+ m[14] * static_cast<T>(a.m[11]);
result.m[11] = m[3] * static_cast<T>(a.m[8])
+ m[7] * static_cast<T>(a.m[9])
+ m[11] * static_cast<T>(a.m[10]);
+ m[15] * static_cast<T>(a.m[11]);
result.m[12] = m[0] * static_cast<T>(a.m[12])
+ m[4] * static_cast<T>(a.m[13])
+ m[8] * static_cast<T>(a.m[14]);
+ m[12] * static_cast<T>(a.m[15]);
result.m[13] = m[1] * static_cast<T>(a.m[12])
+ m[5] * static_cast<T>(a.m[13])
+ m[9] * static_cast<T>(a.m[14]);
+ m[13] * static_cast<T>(a.m[15]);
result.m[14] = m[2] * static_cast<T>(a.m[12])
+ m[6] * static_cast<T>(a.m[13])
+ m[10] * static_cast<T>(a.m[14]);
+ m[14] * static_cast<T>(a.m[15]);
result.m[15] = m[3] * static_cast<T>(a.m[12])
+ m[7] * static_cast<T>(a.m[13])
+ m[11] * static_cast<T>(a.m[14]);
+ m[15] * static_cast<T>(a.m[15]);
m = result.m;
return *this;
}
*/
/**
* matrix vector multiplication
*/
Vector3Template<T> operator * ( const Vector3Template<T> &a )
{
Vector3Template<T> result;
result.x = m[0] * a.x + m[4] * a.y + m[8] * a.z + m[12];
result.y = m[1] * a.x + m[5] * a.y + m[9] * a.z + m[13];
result.z = m[2] * a.x + m[6] * a.y + m[11] * a.z + m[14];
return result;
}
T *data()
{
return &m[0];
}
};
typedef Matrix4x4Template<float> Matrix4x4Float;
typedef Matrix4x4Template<double> Matrix4x4Double;
typedef Matrix4x4Template<Scalar> Matrix4x4;
} // namespace bc
#endif // BLUECORE_MATRIX_H