#ifndef MTKMATH_H_
#define MTKMATH_H_
#include <cmath>
#include <boost/math/tools/precision.hpp>
#include "../types/vect.hpp"
#ifndef M_PI
#define M_PI 3.1415926535897932384626433832795
#endif
namespace MTK {
namespace internal {
template<class Manifold>
struct traits {
typedef typename Manifold::scalar scalar;
enum {DOF = Manifold::DOF};
typedef vect<DOF, scalar> vectorized_type;
typedef Eigen::Matrix<scalar, DOF, DOF> matrix_type;
};
template<>
struct traits<float> : traits<Scalar<float> > {};
template<>
struct traits<double> : traits<Scalar<double> > {};
} // namespace internal
/**
* \defgroup MTKMath Mathematical helper functions
*/
//@{
//! constant @f$ \pi @f$
const double pi = M_PI;
template<class scalar> inline scalar tolerance();
template<> inline float tolerance<float >() { return 1e-5f; }
template<> inline double tolerance<double>() { return 1e-11; }
/**
* normalize @a x to @f$[-bound, bound] @f$.
*
* result for @f$ x = bound + 2\cdot n\cdot bound @f$ is arbitrary @f$\pm bound @f$.
*/
template<class scalar>
inline scalar normalize(scalar x, scalar bound){ //not used
if(std::fabs(x) <= bound) return x;
int r = (int)(x *(scalar(1.0)/ bound));
return x - ((r + (r>>31) + 1) & ~1)*bound;
}
/**
* Calculate cosine and sinc of sqrt(x2).
* @param x2 the squared angle must be non-negative
* @return a pair containing cos and sinc of sqrt(x2)
*/
template<class scalar>
std::pair<scalar, scalar> cos_sinc_sqrt(const scalar &x2){
using std::sqrt;
using std::cos;
using std::sin;
static scalar const taylor_0_bound = boost::math::tools::epsilon<scalar>();
static scalar const taylor_2_bound = sqrt(taylor_0_bound);
static scalar const taylor_n_bound = sqrt(taylor_2_bound);
assert(x2>=0 && "argument must be non-negative");
// FIXME check if bigger bounds are possible
if(x2>=taylor_n_bound) {
// slow fall-back solution
scalar x = sqrt(x2);
return std::make_pair(cos(x), sin(x)/x); // x is greater than 0.
}
// FIXME Replace by Horner-Scheme (4 instead of 5 FLOP/term, numerically more stable, theoretically cos and sinc can be calculated in parallel using SSE2 mulpd/addpd)
// TODO Find optimal coefficients using Remez algorithm
static scalar const inv[] = {1/3., 1/4., 1/5., 1/6., 1/7., 1/8., 1/9.};
scalar cosi = 1., sinc=1;
scalar term = -1/2. * x2;
for(int i=0; i<3; ++i) {
cosi += term;
term *= inv[2*i];
sinc += term;
term *= -inv[2*i+1] * x2;
}
return std::make_pair(cosi, sinc);
}
template<typename Base>
Eigen::Matrix<typename Base::scalar, 3, 3> hat(const Base& v) {
Eigen::Matrix<typename Base::scalar, 3, 3> res;
res << 0, -v[2], v[1],
v[2], 0, -v[0],
-v[1], v[0], 0;
return res;
}
template<typename Base>
Eigen::Matrix<typename Base::scalar, 3, 3> A_inv_trans(const Base& v){
Eigen::Matrix<typename Base::scalar, 3, 3> res;
if(v.norm() > MTK::tolerance<typename Base::scalar>())
{
res = Eigen::Matrix<typename Base::scalar, 3, 3>::Identity() + 0.5 * hat<Base>(v) + (1 - v.norm() * std::cos(v.norm() / 2) / 2 / std::sin(v.norm() / 2)) * hat(v) * hat(v) / v.squaredNorm();
}
else
{
res = Eigen::Matrix<typename Base::scalar, 3, 3>::Identity();
}
return res;
}
template<typename Base>
Eigen::Matrix<typename Base::scalar, 3, 3> A_inv(const Base& v){
Eigen::Matrix<typename Base::scalar, 3, 3> res;
if(v.norm() > MTK::tolerance<typename Base::scalar>())
{
res = Eigen::Matrix<typename Base::scalar, 3, 3>::Identity() - 0.5 * hat<Base>(v) + (1 - v.norm() * std::cos(v.norm() / 2) / 2 / std::sin(v.norm() / 2)) * hat(v) * hat(v) / v.squaredNorm();
}
else
{
res = Eigen::Matrix<typename Base::scalar, 3, 3>::Identity();
}
return res;
}
template<typename scalar>
Eigen::Matrix<scalar, 2, 3> S2_w_expw_( Eigen::Matrix<scalar, 2, 1> v, scalar length)
{
Eigen::Matrix<scalar, 2, 3> res;
scalar norm = std::sqrt(v[0]*v[0] + v[1]*v[1]);
if(norm < MTK::tolerance<scalar>()){
res = Eigen::Matrix<scalar, 2, 3>::Zero();
res(0, 1) = 1;
res(1, 2) = 1;
res /= length;
}
else{
res << -v[0]*(1/norm-1/std::tan(norm))/std::sin(norm), norm/std::sin(norm), 0,
-v[1]*(1/norm-1/std::tan(norm))/std::sin(norm), 0, norm/std::sin(norm);
res /= length;
}
}
template<typename Base>
Eigen::Matrix<typename Base::scalar, 3, 3> A_matrix(const Base & v){
Eigen::Matrix<typename Base::scalar, 3, 3> res;
double squaredNorm = v[0] * v[0] + v[1] * v[1] + v[2] * v[2];
double norm = std::sqrt(squaredNorm);
if(norm < MTK::tolerance<typename Base::scalar>()){
res = Eigen::Matrix<typename Base::scalar, 3, 3>::Identity();
}
else{
res = Eigen::Matrix<typename Base::scalar, 3, 3>::Identity() + (1 - std::cos(norm)) / squaredNorm * hat(v) + (1 - std::sin(norm) / norm) / squaredNorm * hat(v) * hat(v);
}
return res;
}
template<class scalar, int n>
scalar exp(vectview<scalar, n> result, vectview<const scalar, n> vec, const scalar& scale = 1) {
scalar norm2 = vec.squaredNorm();
std::pair<scalar, scalar> cos_sinc = cos_sinc_sqrt(scale*scale * norm2);
scalar mult = cos_sinc.second * scale;
result = mult * vec;
return cos_sinc.first;
}
/**
* Inverse function to @c exp.
*
* @param result @c vectview to the result
* @param w scalar part of input
* @param vec vector part of input
* @param scale scale result by this value
* @param plus_minus_periodicity if true values @f$[w, vec]@f$ and @f$[-w, -vec]@f$ give the same result
*/
template<class scalar, int n>
void log(vectview<scalar, n> result,
const scalar &w, const vectview<const scalar, n> vec,
const scalar &scale, bool plus_minus_periodicity)
{
// FIXME implement optimized case for vec.squaredNorm() <= tolerance() * (w*w) via Rational Remez approximation ~> only one division
scalar nv = vec.norm();
if(nv < tolerance<scalar>()) {
if(!plus_minus_periodicity && w < 0) {
// find the maximal entry:
int i;
nv = vec.cwiseAbs().maxCoeff(&i);
result = scale * std::atan2(nv, w) * vect<n, scalar>::Unit(i);
return;
}
nv = tolerance<scalar>();
}
scalar s = scale / nv * (plus_minus_periodicity ? std::atan(nv / w) : std::atan2(nv, w) );
result = s * vec;
}
} // namespace MTK
#endif /* MTKMATH_H_ */
详细注释上述代码
最新发布