devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

merge with main repository

Browse Source
This commit is contained in:
Thomas Capricelli
2009-11-08 22:27:32 +01:00
parent 456f7d094d aa0974286f
commit 751a333491
65 changed files with 2753 additions and 476 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
unsupported/Eigen/src/AutoDiff/AutoDiffJacobian.h
View File

@@ -46,13 +46,16 @@ public:
InputsAtCompileTime = Functor::InputsAtCompileTime,
ValuesAtCompileTime = Functor::ValuesAtCompileTime
};
typedef typename Functor::InputType InputType;
typedef typename Functor::ValueType ValueType;
typedef typename Functor::JacobianType JacobianType;
typedef typename JacobianType::Scalar Scalar;
typedef Matrix<Scalar,InputsAtCompileTime,1> DerivativeType;
typedef AutoDiffScalar<DerivativeType> ActiveScalar;
typedef AutoDiffScalar<Matrix<double,InputsAtCompileTime,1> > ActiveScalar;
typedef Matrix<ActiveScalar, InputsAtCompileTime, 1> ActiveInput;
typedef Matrix<ActiveScalar, ValuesAtCompileTime, 1> ActiveValue;
@@ -69,26 +72,20 @@ public:
ActiveInput ax = x.template cast<ActiveScalar>();
ActiveValue av(jac.rows());
if(InputsAtCompileTime==Dynamic)
{
for (int j=0; j<jac.cols(); j++)
ax[j].derivatives().resize(this->inputs());
for (int j=0; j<jac.rows(); j++)
av[j].derivatives().resize(this->inputs());
}
for (int j=0; j<jac.cols(); j++)
for (int i=0; i<jac.cols(); i++)
ax[i].derivatives().coeffRef(j) = i==j ? 1 : 0;
for (int i=0; i<jac.cols(); i++)
ax[i].derivatives() = DerivativeType::Unit(this->inputs(),i);
Functor::operator()(ax, &av);
for (int i=0; i<jac.rows(); i++)
{
(*v)[i] = av[i].value();
for (int j=0; j<jac.cols(); j++)
jac.coeffRef(i,j) = av[i].derivatives().coeff(j);
jac.row(i) = av[i].derivatives();
}
}
protected:

225
unsupported/Eigen/src/AutoDiff/AutoDiffScalar.h
View File

@@ -27,15 +27,35 @@
namespace Eigen {
template<typename A, typename B>
struct ei_make_coherent_impl {
static void run(A& a, B& b) {}
};
// resize a to match b is a.size()==0, and conversely.
template<typename A, typename B>
void ei_make_coherent(const A& a, const B&b)
{
ei_make_coherent_impl<A,B>::run(a.const_cast_derived(), b.const_cast_derived());
}
/** \class AutoDiffScalar
* \brief A scalar type replacement with automatic differentation capability
*
* \param DerType the vector type used to store/represent the derivatives (e.g. Vector3f)
* \param _DerType the vector type used to store/represent the derivatives. The base scalar type
* as well as the number of derivatives to compute are determined from this type.
* Typical choices include, e.g., \c Vector4f for 4 derivatives, or \c VectorXf
* if the number of derivatives is not known at compile time, and/or, the number
* of derivatives is large.
* Note that _DerType can also be a reference (e.g., \c VectorXf&) to wrap a
* existing vector into an AutoDiffScalar.
* Finally, _DerType can also be any Eigen compatible expression.
*
* This class represents a scalar value while tracking its respective derivatives.
* This class represents a scalar value while tracking its respective derivatives using Eigen's expression
* template mechanism.
*
* It supports the following list of global math function:
* - std::abs, std::sqrt, std::pow, std::exp, std::log, std::sin, std::cos,
* - std::abs, std::sqrt, std::pow, std::exp, std::log, std::sin, std::cos,
* - ei_abs, ei_sqrt, ei_pow, ei_exp, ei_log, ei_sin, ei_cos,
* - ei_conj, ei_real, ei_imag, ei_abs2.
*
@@ -44,34 +64,35 @@ namespace Eigen {
* while derivatives are computed right away.
*
*/
template<typename DerType>
template<typename _DerType>
class AutoDiffScalar
{
public:
typedef typename ei_cleantype<_DerType>::type DerType;
typedef typename ei_traits<DerType>::Scalar Scalar;
inline AutoDiffScalar() {}
inline AutoDiffScalar(const Scalar& value)
: m_value(value)
{
if(m_derivatives.size()>0)
m_derivatives.setZero();
}
inline AutoDiffScalar(const Scalar& value, const DerType& der)
: m_value(value), m_derivatives(der)
{}
template<typename OtherDerType>
inline AutoDiffScalar(const AutoDiffScalar<OtherDerType>& other)
: m_value(other.value()), m_derivatives(other.derivatives())
{}
inline AutoDiffScalar(const AutoDiffScalar& other)
: m_value(other.value()), m_derivatives(other.derivatives())
{}
template<typename OtherDerType>
inline AutoDiffScalar& operator=(const AutoDiffScalar<OtherDerType>& other)
{
@@ -79,32 +100,49 @@ class AutoDiffScalar
m_derivatives = other.derivatives();
return *this;
}
inline AutoDiffScalar& operator=(const AutoDiffScalar& other)
{
m_value = other.value();
m_derivatives = other.derivatives();
return *this;
}
// inline operator const Scalar& () const { return m_value; }
// inline operator Scalar& () { return m_value; }
inline const Scalar& value() const { return m_value; }
inline Scalar& value() { return m_value; }
inline const DerType& derivatives() const { return m_derivatives; }
inline DerType& derivatives() { return m_derivatives; }
inline const AutoDiffScalar<DerType&> operator+(const Scalar& other) const
{
return AutoDiffScalar<DerType>(m_value + other, m_derivatives);
}
friend inline const AutoDiffScalar<DerType&> operator+(const Scalar& a, const AutoDiffScalar& b)
{
return AutoDiffScalar<DerType>(a + b.value(), b.derivatives());
}
inline AutoDiffScalar& operator+=(const Scalar& other)
{
value() += other;
return *this;
}
template<typename OtherDerType>
inline const AutoDiffScalar<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerType,OtherDerType> >
inline const AutoDiffScalar<typename MakeCwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerType,typename ei_cleantype<OtherDerType>::type>::Type >
operator+(const AutoDiffScalar<OtherDerType>& other) const
{
return AutoDiffScalar<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerType,OtherDerType> >(
ei_make_coherent(m_derivatives, other.derivatives());
return AutoDiffScalar<typename MakeCwiseBinaryOp<ei_scalar_sum_op<Scalar>,DerType,typename ei_cleantype<OtherDerType>::type>::Type >(
m_value + other.value(),
m_derivatives + other.derivatives());
}
template<typename OtherDerType>
inline AutoDiffScalar&
operator+=(const AutoDiffScalar<OtherDerType>& other)
@@ -112,16 +150,17 @@ class AutoDiffScalar
(*this) = (*this) + other;
return *this;
}
template<typename OtherDerType>
inline const AutoDiffScalar<CwiseBinaryOp<ei_scalar_difference_op<Scalar>, DerType,OtherDerType> >
inline const AutoDiffScalar<typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>, DerType,typename ei_cleantype<OtherDerType>::type>::Type >
operator-(const AutoDiffScalar<OtherDerType>& other) const
{
return AutoDiffScalar<CwiseBinaryOp<ei_scalar_difference_op<Scalar>, DerType,OtherDerType> >(
ei_make_coherent(m_derivatives, other.derivatives());
return AutoDiffScalar<typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>, DerType,typename ei_cleantype<OtherDerType>::type>::Type >(
m_value - other.value(),
m_derivatives - other.derivatives());
}
template<typename OtherDerType>
inline AutoDiffScalar&
operator-=(const AutoDiffScalar<OtherDerType>& other)
@@ -129,104 +168,151 @@ class AutoDiffScalar
*this = *this - other;
return *this;
}
template<typename OtherDerType>
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, DerType> >
inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_opposite_op<Scalar>, DerType>::Type >
operator-() const
{
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, DerType> >(
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_opposite_op<Scalar>, DerType>::Type >(
-m_value,
-m_derivatives);
}
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >
inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >
operator*(const Scalar& other) const
{
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >(
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >(
m_value * other,
(m_derivatives * other));
}
friend inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >
friend inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >
operator*(const Scalar& other, const AutoDiffScalar& a)
{
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >(
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >(
a.value() * other,
a.derivatives() * other);
}
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >
inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >
operator/(const Scalar& other) const
{
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >(
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >(
m_value / other,
(m_derivatives * (Scalar(1)/other)));
}
friend inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >
friend inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >
operator/(const Scalar& other, const AutoDiffScalar& a)
{
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >(
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >(
other / a.value(),
a.derivatives() * (-Scalar(1)/other));
}
template<typename OtherDerType>
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>,
NestByValue<CwiseBinaryOp<ei_scalar_difference_op<Scalar>,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, OtherDerType> > > > > >
inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>,
typename MakeNestByValue<typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type>::Type,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, typename ei_cleantype<OtherDerType>::type>::Type>::Type >::Type >::Type >::Type >
operator/(const AutoDiffScalar<OtherDerType>& other) const
{
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>,
NestByValue<CwiseBinaryOp<ei_scalar_difference_op<Scalar>,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, OtherDerType> > > > > >(
ei_make_coherent(m_derivatives, other.derivatives());
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>,
typename MakeNestByValue<typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type>::Type,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, typename ei_cleantype<OtherDerType>::type>::Type>::Type >::Type >::Type >::Type >(
m_value / other.value(),
((m_derivatives * other.value()).nestByValue() - (m_value * other.derivatives()).nestByValue()).nestByValue()
* (Scalar(1)/(other.value()*other.value())));
}
template<typename OtherDerType>
inline const AutoDiffScalar<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, OtherDerType> > > >
inline const AutoDiffScalar<typename MakeCwiseBinaryOp<ei_scalar_sum_op<Scalar>,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type>::Type,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, typename ei_cleantype<OtherDerType>::type>::Type>::Type >::Type >
operator*(const AutoDiffScalar<OtherDerType>& other) const
{
return AutoDiffScalar<CwiseBinaryOp<ei_scalar_sum_op<Scalar>,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >,
NestByValue<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, OtherDerType> > > >(
ei_make_coherent(m_derivatives, other.derivatives());
return AutoDiffScalar<typename MakeCwiseBinaryOp<ei_scalar_sum_op<Scalar>,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type>::Type,
typename MakeNestByValue<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, typename ei_cleantype<OtherDerType>::type>::Type>::Type >::Type >(
m_value * other.value(),
(m_derivatives * other.value()).nestByValue() + (m_value * other.derivatives()).nestByValue());
}
inline AutoDiffScalar& operator*=(const Scalar& other)
{
*this = *this * other;
return *this;
}
template<typename OtherDerType>
inline AutoDiffScalar& operator*=(const AutoDiffScalar<OtherDerType>& other)
{
*this = *this * other;
return *this;
}
protected:
Scalar m_value;
DerType m_derivatives;
};
template<typename A_Scalar, int A_Rows, int A_Cols, int A_Options, int A_MaxRows, int A_MaxCols, typename B>
struct ei_make_coherent_impl<Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols>, B> {
typedef Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols> A;
static void run(A& a, B& b) {
if((A_Rows==Dynamic || A_Cols==Dynamic) && (a.size()==0))
{
a.resize(b.size());
a.setZero();
}
}
};
template<typename A, typename B_Scalar, int B_Rows, int B_Cols, int B_Options, int B_MaxRows, int B_MaxCols>
struct ei_make_coherent_impl<A, Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> > {
typedef Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> B;
static void run(A& a, B& b) {
if((B_Rows==Dynamic || B_Cols==Dynamic) && (b.size()==0))
{
b.resize(a.size());
b.setZero();
}
}
};
template<typename A_Scalar, int A_Rows, int A_Cols, int A_Options, int A_MaxRows, int A_MaxCols,
typename B_Scalar, int B_Rows, int B_Cols, int B_Options, int B_MaxRows, int B_MaxCols>
struct ei_make_coherent_impl<Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols>,
Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> > {
typedef Matrix<A_Scalar, A_Rows, A_Cols, A_Options, A_MaxRows, A_MaxCols> A;
typedef Matrix<B_Scalar, B_Rows, B_Cols, B_Options, B_MaxRows, B_MaxCols> B;
static void run(A& a, B& b) {
if((A_Rows==Dynamic || A_Cols==Dynamic) && (a.size()==0))
{
a.resize(b.size());
a.setZero();
}
else if((B_Rows==Dynamic || B_Cols==Dynamic) && (b.size()==0))
{
b.resize(a.size());
b.setZero();
}
}
};
}
#define EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(FUNC,CODE) \
template<typename DerType> \
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<typename ei_traits<DerType>::Scalar>, DerType> > \
FUNC(const AutoDiffScalar<DerType>& x) { \
inline const Eigen::AutoDiffScalar<typename Eigen::MakeCwiseUnaryOp<Eigen::ei_scalar_multiple_op<typename Eigen::ei_traits<DerType>::Scalar>, DerType>::Type > \
FUNC(const Eigen::AutoDiffScalar<DerType>& x) { \
using namespace Eigen; \
typedef typename ei_traits<DerType>::Scalar Scalar; \
typedef AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> > ReturnType; \
typedef AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type > ReturnType; \
CODE; \
}
@@ -234,34 +320,35 @@ namespace std
{
EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(abs,
return ReturnType(std::abs(x.value()), x.derivatives() * (sign(x.value())));)
EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(sqrt,
Scalar sqrtx = std::sqrt(x.value());
return ReturnType(sqrtx,x.derivatives() * (Scalar(0.5) / sqrtx));)
EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(cos,
return ReturnType(std::cos(x.value()), x.derivatives() * (-std::sin(x.value())));)
EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(sin,
return ReturnType(std::sin(x.value()),x.derivatives() * std::cos(x.value()));)
EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(exp,
Scalar expx = std::exp(x.value());
return ReturnType(expx,x.derivatives() * expx);)
EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(ei_log,
return ReturnType(std::log(x.value),x.derivatives() * (Scalar(1).x.value()));)
template<typename DerType>
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<typename ei_traits<DerType>::Scalar>, DerType> >
pow(const AutoDiffScalar<DerType>& x, typename ei_traits<DerType>::Scalar y)
inline const Eigen::AutoDiffScalar<typename Eigen::MakeCwiseUnaryOp<Eigen::ei_scalar_multiple_op<typename Eigen::ei_traits<DerType>::Scalar>, DerType>::Type >
pow(const Eigen::AutoDiffScalar<DerType>& x, typename Eigen::ei_traits<DerType>::Scalar y)
{
using namespace Eigen;
typedef typename ei_traits<DerType>::Scalar Scalar;
return AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType> >(
return AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, DerType>::Type >(
std::pow(x.value(),y),
x.derivatives() * (y * std::pow(x.value(),y-1)));
}
}
namespace Eigen {
@@ -297,7 +384,7 @@ EIGEN_AUTODIFF_DECLARE_GLOBAL_UNARY(ei_log,
return ReturnType(ei_log(x.value),x.derivatives() * (Scalar(1).x.value()));)
template<typename DerType>
inline const AutoDiffScalar<CwiseUnaryOp<ei_scalar_multiple_op<typename ei_traits<DerType>::Scalar>, DerType> >
inline const AutoDiffScalar<typename MakeCwiseUnaryOp<ei_scalar_multiple_op<typename ei_traits<DerType>::Scalar>, DerType>::Type >
ei_pow(const AutoDiffScalar<DerType>& x, typename ei_traits<DerType>::Scalar y)
{ return std::pow(x,y);}

128
unsupported/Eigen/src/AutoDiff/AutoDiffVector.h
View File

@@ -35,7 +35,7 @@ namespace Eigen {
* This class represents a scalar value while tracking its respective derivatives.
*
* It supports the following list of global math function:
* - std::abs, std::sqrt, std::pow, std::exp, std::log, std::sin, std::cos,
* - std::abs, std::sqrt, std::pow, std::exp, std::log, std::sin, std::cos,
* - ei_abs, ei_sqrt, ei_pow, ei_exp, ei_log, ei_sin, ei_cos,
* - ei_conj, ei_real, ei_imag, ei_abs2.
*
@@ -48,130 +48,150 @@ template<typename ValueType, typename JacobianType>
class AutoDiffVector
{
public:
typedef typename ei_traits<ValueType>::Scalar Scalar;
//typedef typename ei_traits<ValueType>::Scalar Scalar;
typedef typename ei_traits<ValueType>::Scalar BaseScalar;
typedef AutoDiffScalar<Matrix<BaseScalar,JacobianType::RowsAtCompileTime,1> > ActiveScalar;
typedef ActiveScalar Scalar;
typedef AutoDiffScalar<typename JacobianType::ColXpr> CoeffType;
inline AutoDiffVector() {}
inline AutoDiffVector(const ValueType& values)
: m_values(values)
{
m_jacobian.setZero();
}
CoeffType operator[] (int i) { return CoeffType(m_values[i], m_jacobian.col(i)); }
const CoeffType operator[] (int i) const { return CoeffType(m_values[i], m_jacobian.col(i)); }
CoeffType operator() (int i) { return CoeffType(m_values[i], m_jacobian.col(i)); }
const CoeffType operator() (int i) const { return CoeffType(m_values[i], m_jacobian.col(i)); }
CoeffType coeffRef(int i) { return CoeffType(m_values[i], m_jacobian.col(i)); }
const CoeffType coeffRef(int i) const { return CoeffType(m_values[i], m_jacobian.col(i)); }
int size() const { return m_values.size(); }
// FIXME here we could return an expression of the sum
Scalar sum() const { /*std::cerr << "sum \n\n";*/ /*std::cerr << m_jacobian.rowwise().sum() << "\n\n";*/ return Scalar(m_values.sum(), m_jacobian.rowwise().sum()); }
inline AutoDiffVector(const ValueType& values, const JacobianType& jac)
: m_values(values), m_jacobian(jac)
{}
template<typename OtherValueType, typename OtherJacobianType>
inline AutoDiffVector(const AutoDiffVector<OtherValueType, OtherJacobianType>& other)
: m_values(other.values()), m_jacobian(other.jacobian())
{}
inline AutoDiffVector(const AutoDiffVector& other)
: m_values(other.values()), m_jacobian(other.jacobian())
{}
template<typename OtherValueType, typename OtherJacobianType>
inline AutoDiffScalar& operator=(const AutoDiffVector<OtherValueType, OtherJacobianType>& other)
inline AutoDiffVector& operator=(const AutoDiffVector<OtherValueType, OtherJacobianType>& other)
{
m_values = other.values();
m_jacobian = other.jacobian();
return *this;
}
inline AutoDiffVector& operator=(const AutoDiffVector& other)
{
m_values = other.values();
m_jacobian = other.jacobian();
return *this;
}
inline const ValueType& values() const { return m_values; }
inline ValueType& values() { return m_values; }
inline const JacobianType& jacobian() const { return m_jacobian; }
inline JacobianType& jacobian() { return m_jacobian; }
template<typename OtherValueType,typename OtherJacobianType>
inline const AutoDiffVector<
CwiseBinaryOp<ei_scalar_sum_op<Scalar>,ValueType,OtherValueType> >
CwiseBinaryOp<ei_scalar_sum_op<Scalar>,JacobianType,OtherJacobianType> >
operator+(const AutoDiffScalar<OtherDerType>& other) const
typename MakeCwiseBinaryOp<ei_scalar_sum_op<BaseScalar>,ValueType,OtherValueType>::Type,
typename MakeCwiseBinaryOp<ei_scalar_sum_op<BaseScalar>,JacobianType,OtherJacobianType>::Type >
operator+(const AutoDiffVector<OtherValueType,OtherJacobianType>& other) const
{
return AutoDiffVector<
CwiseBinaryOp<ei_scalar_sum_op<Scalar>,ValueType,OtherValueType> >
CwiseBinaryOp<ei_scalar_sum_op<Scalar>,JacobianType,OtherJacobianType> >(
typename MakeCwiseBinaryOp<ei_scalar_sum_op<BaseScalar>,ValueType,OtherValueType>::Type,
typename MakeCwiseBinaryOp<ei_scalar_sum_op<BaseScalar>,JacobianType,OtherJacobianType>::Type >(
m_values + other.values(),
m_jacobian + other.jacobian());
}
template<typename OtherValueType, typename OtherJacobianType>
inline AutoDiffVector&
operator+=(const AutoDiffVector<OtherValueType,OtherDerType>& other)
operator+=(const AutoDiffVector<OtherValueType,OtherJacobianType>& other)
{
m_values += other.values();
m_jacobian += other.jacobian();
return *this;
}
template<typename OtherValueType,typename OtherJacobianType>
inline const AutoDiffVector<
CwiseBinaryOp<ei_scalar_difference_op<Scalar>,ValueType,OtherValueType> >
CwiseBinaryOp<ei_scalar_difference_op<Scalar>,JacobianType,OtherJacobianType> >
operator-(const AutoDiffScalar<OtherDerType>& other) const
typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>,ValueType,OtherValueType>::Type,
typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>,JacobianType,OtherJacobianType>::Type >
operator-(const AutoDiffVector<OtherValueType,OtherJacobianType>& other) const
{
return AutoDiffVector<
CwiseBinaryOp<ei_scalar_difference_op<Scalar>,ValueType,OtherValueType> >
CwiseBinaryOp<ei_scalar_difference_op<Scalar>,JacobianType,OtherJacobianType> >(
m_values - other.values(),
m_jacobian - other.jacobian());
typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>,ValueType,OtherValueType>::Type,
typename MakeCwiseBinaryOp<ei_scalar_difference_op<Scalar>,JacobianType,OtherJacobianType>::Type >(
m_values - other.values(),
m_jacobian - other.jacobian());
}
template<typename OtherValueType, typename OtherJacobianType>
inline AutoDiffVector&
operator-=(const AutoDiffVector<OtherValueType,OtherDerType>& other)
operator-=(const AutoDiffVector<OtherValueType,OtherJacobianType>& other)
{
m_values -= other.values();
m_jacobian -= other.jacobian();
return *this;
}
inline const AutoDiffVector<
CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, ValueType>
CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, JacobianType> >
typename MakeCwiseUnaryOp<ei_scalar_opposite_op<Scalar>, ValueType>::Type,
typename MakeCwiseUnaryOp<ei_scalar_opposite_op<Scalar>, JacobianType>::Type >
operator-() const
{
return AutoDiffVector<
CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, ValueType>
CwiseUnaryOp<ei_scalar_opposite_op<Scalar>, JacobianType> >(
-m_values,
-m_jacobian);
typename MakeCwiseUnaryOp<ei_scalar_opposite_op<Scalar>, ValueType>::Type,
typename MakeCwiseUnaryOp<ei_scalar_opposite_op<Scalar>, JacobianType>::Type >(
-m_values,
-m_jacobian);
}
inline const AutoDiffVector<
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType> >
operator*(const Scalar& other) const
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>::Type,
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType>::Type>
operator*(const BaseScalar& other) const
{
return AutoDiffVector<
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType> >(
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>::Type,
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType>::Type >(
m_values * other,
(m_jacobian * other));
m_jacobian * other);
}
friend inline const AutoDiffVector<
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType> >
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>::Type,
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType>::Type >
operator*(const Scalar& other, const AutoDiffVector& v)
{
return AutoDiffVector<
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>
CwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType> >(
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, ValueType>::Type,
typename MakeCwiseUnaryOp<ei_scalar_multiple_op<Scalar>, JacobianType>::Type >(
v.values() * other,
v.jacobian() * other);
}
// template<typename OtherValueType,typename OtherJacobianType>
// inline const AutoDiffVector<
// CwiseBinaryOp<ei_scalar_multiple_op<Scalar>, ValueType, OtherValueType>
@@ -188,25 +208,25 @@ class AutoDiffVector
// m_values.cwise() * other.values(),
// (m_jacobian * other.values()).nestByValue() + (m_values * other.jacobian()).nestByValue());
// }
inline AutoDiffVector& operator*=(const Scalar& other)
{
m_values *= other;
m_jacobian *= other;
return *this;
}
template<typename OtherValueType,typename OtherJacobianType>
inline AutoDiffVector& operator*=(const AutoDiffVector<OtherValueType,OtherJacobianType>& other)
{
*this = *this * other;
return *this;
}
protected:
ValueType m_values;
JacobianType m_jacobian;
};
}

213
unsupported/Eigen/src/FFT/ei_fftw_impl.h Normal file
View File

@@ -0,0 +1,213 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2009 Mark Borgerding mark a borgerding net
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
// FFTW uses non-const arguments
// so we must use ugly const_cast calls for all the args it uses
//
// This should be safe as long as
// 1. we use FFTW_ESTIMATE for all our planning
// see the FFTW docs section 4.3.2 "Planner Flags"
// 2. fftw_complex is compatible with std::complex
// This assumes std::complex<T> layout is array of size 2 with real,imag
template <typename T>
inline
T * ei_fftw_cast(const T* p)
{
return const_cast<T*>( p);
}
inline
fftw_complex * ei_fftw_cast( const std::complex<double> * p)
{
return const_cast<fftw_complex*>( reinterpret_cast<const fftw_complex*>(p) );
}
inline
fftwf_complex * ei_fftw_cast( const std::complex<float> * p)
{
return const_cast<fftwf_complex*>( reinterpret_cast<const fftwf_complex*>(p) );
}
inline
fftwl_complex * ei_fftw_cast( const std::complex<long double> * p)
{
return const_cast<fftwl_complex*>( reinterpret_cast<const fftwl_complex*>(p) );
}
template <typename T>
struct ei_fftw_plan {};
template <>
struct ei_fftw_plan<float>
{
typedef float scalar_type;
typedef fftwf_complex complex_type;
fftwf_plan m_plan;
ei_fftw_plan() :m_plan(NULL) {}
~ei_fftw_plan() {if (m_plan) fftwf_destroy_plan(m_plan);}
inline
void fwd(complex_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftwf_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE);
fftwf_execute_dft( m_plan, src,dst);
}
inline
void inv(complex_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftwf_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE);
fftwf_execute_dft( m_plan, src,dst);
}
inline
void fwd(complex_type * dst,scalar_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftwf_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE);
fftwf_execute_dft_r2c( m_plan,src,dst);
}
inline
void inv(scalar_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL)
m_plan = fftwf_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE);
fftwf_execute_dft_c2r( m_plan, src,dst);
}
};
template <>
struct ei_fftw_plan<double>
{
typedef double scalar_type;
typedef fftw_complex complex_type;
fftw_plan m_plan;
ei_fftw_plan() :m_plan(NULL) {}
~ei_fftw_plan() {if (m_plan) fftw_destroy_plan(m_plan);}
inline
void fwd(complex_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftw_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE);
fftw_execute_dft( m_plan, src,dst);
}
inline
void inv(complex_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftw_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE);
fftw_execute_dft( m_plan, src,dst);
}
inline
void fwd(complex_type * dst,scalar_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftw_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE);
fftw_execute_dft_r2c( m_plan,src,dst);
}
inline
void inv(scalar_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL)
m_plan = fftw_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE);
fftw_execute_dft_c2r( m_plan, src,dst);
}
};
template <>
struct ei_fftw_plan<long double>
{
typedef long double scalar_type;
typedef fftwl_complex complex_type;
fftwl_plan m_plan;
ei_fftw_plan() :m_plan(NULL) {}
~ei_fftw_plan() {if (m_plan) fftwl_destroy_plan(m_plan);}
inline
void fwd(complex_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftwl_plan_dft_1d(nfft,src,dst, FFTW_FORWARD, FFTW_ESTIMATE);
fftwl_execute_dft( m_plan, src,dst);
}
inline
void inv(complex_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftwl_plan_dft_1d(nfft,src,dst, FFTW_BACKWARD , FFTW_ESTIMATE);
fftwl_execute_dft( m_plan, src,dst);
}
inline
void fwd(complex_type * dst,scalar_type * src,int nfft) {
if (m_plan==NULL) m_plan = fftwl_plan_dft_r2c_1d(nfft,src,dst,FFTW_ESTIMATE);
fftwl_execute_dft_r2c( m_plan,src,dst);
}
inline
void inv(scalar_type * dst,complex_type * src,int nfft) {
if (m_plan==NULL)
m_plan = fftwl_plan_dft_c2r_1d(nfft,src,dst,FFTW_ESTIMATE);
fftwl_execute_dft_c2r( m_plan, src,dst);
}
};
template <typename _Scalar>
struct ei_fftw_impl
{
typedef _Scalar Scalar;
typedef std::complex<Scalar> Complex;
inline
void clear()
{
m_plans.clear();
}
// complex-to-complex forward FFT
inline
void fwd( Complex * dst,const Complex *src,int nfft)
{
get_plan(nfft,false,dst,src).fwd(ei_fftw_cast(dst), ei_fftw_cast(src),nfft );
}
// real-to-complex forward FFT
inline
void fwd( Complex * dst,const Scalar * src,int nfft)
{
get_plan(nfft,false,dst,src).fwd(ei_fftw_cast(dst), ei_fftw_cast(src) ,nfft);
}
// inverse complex-to-complex
inline
void inv(Complex * dst,const Complex *src,int nfft)
{
get_plan(nfft,true,dst,src).inv(ei_fftw_cast(dst), ei_fftw_cast(src),nfft );
}
// half-complex to scalar
inline
void inv( Scalar * dst,const Complex * src,int nfft)
{
get_plan(nfft,true,dst,src).inv(ei_fftw_cast(dst), ei_fftw_cast(src),nfft );
}
protected:
typedef ei_fftw_plan<Scalar> PlanData;
typedef std::map<int,PlanData> PlanMap;
PlanMap m_plans;
inline
PlanData & get_plan(int nfft,bool inverse,void * dst,const void * src)
{
bool inplace = (dst==src);
bool aligned = ( (reinterpret_cast<size_t>(src)&15) | (reinterpret_cast<size_t>(dst)&15) ) == 0;
int key = (nfft<<3 ) | (inverse<<2) | (inplace<<1) | aligned;
return m_plans[key];
}
};
/* vim: set filetype=cpp et sw=2 ts=2 ai: */

410
unsupported/Eigen/src/FFT/ei_kissfft_impl.h Normal file
View File

@@ -0,0 +1,410 @@
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2009 Mark Borgerding mark a borgerding net
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
// This FFT implementation was derived from kissfft http:sourceforge.net/projects/kissfft
// Copyright 2003-2009 Mark Borgerding
template <typename _Scalar>
struct ei_kiss_cpx_fft
{
typedef _Scalar Scalar;
typedef std::complex<Scalar> Complex;
std::vector<Complex> m_twiddles;
std::vector<int> m_stageRadix;
std::vector<int> m_stageRemainder;
std::vector<Complex> m_scratchBuf;
bool m_inverse;
inline
void make_twiddles(int nfft,bool inverse)
{
m_inverse = inverse;
m_twiddles.resize(nfft);
Scalar phinc = (inverse?2:-2)* acos( (Scalar) -1) / nfft;
for (int i=0;i<nfft;++i)
m_twiddles[i] = exp( Complex(0,i*phinc) );
}
void factorize(int nfft)
{
//start factoring out 4's, then 2's, then 3,5,7,9,...
int n= nfft;
int p=4;
do {
while (n % p) {
switch (p) {
case 4: p = 2; break;
case 2: p = 3; break;
default: p += 2; break;
}
if (p*p>n)
p=n;// impossible to have a factor > sqrt(n)
}
n /= p;
m_stageRadix.push_back(p);
m_stageRemainder.push_back(n);
if ( p > 5 )
m_scratchBuf.resize(p); // scratchbuf will be needed in bfly_generic
}while(n>1);
}
template <typename _Src>
inline
void work( int stage,Complex * xout, const _Src * xin, size_t fstride,size_t in_stride)
{
int p = m_stageRadix[stage];
int m = m_stageRemainder[stage];
Complex * Fout_beg = xout;
Complex * Fout_end = xout + p*m;
if (m>1) {
do{
// recursive call:
// DFT of size m*p performed by doing
// p instances of smaller DFTs of size m,
// each one takes a decimated version of the input
work(stage+1, xout , xin, fstride*p,in_stride);
xin += fstride*in_stride;
}while( (xout += m) != Fout_end );
}else{
do{
*xout = *xin;
xin += fstride*in_stride;
}while(++xout != Fout_end );
}
xout=Fout_beg;
// recombine the p smaller DFTs
switch (p) {
case 2: bfly2(xout,fstride,m); break;
case 3: bfly3(xout,fstride,m); break;
case 4: bfly4(xout,fstride,m); break;
case 5: bfly5(xout,fstride,m); break;
default: bfly_generic(xout,fstride,m,p); break;
}
}
inline
void bfly2( Complex * Fout, const size_t fstride, int m)
{
for (int k=0;k<m;++k) {
Complex t = Fout[m+k] * m_twiddles[k*fstride];
Fout[m+k] = Fout[k] - t;
Fout[k] += t;
}
}
inline
void bfly4( Complex * Fout, const size_t fstride, const size_t m)
{
Complex scratch[6];
int negative_if_inverse = m_inverse * -2 +1;
for (size_t k=0;k<m;++k) {
scratch[0] = Fout[k+m] * m_twiddles[k*fstride];
scratch[1] = Fout[k+2*m] * m_twiddles[k*fstride*2];
scratch[2] = Fout[k+3*m] * m_twiddles[k*fstride*3];
scratch[5] = Fout[k] - scratch[1];
Fout[k] += scratch[1];
scratch[3] = scratch[0] + scratch[2];
scratch[4] = scratch[0] - scratch[2];
scratch[4] = Complex( scratch[4].imag()*negative_if_inverse , -scratch[4].real()* negative_if_inverse );
Fout[k+2*m] = Fout[k] - scratch[3];
Fout[k] += scratch[3];
Fout[k+m] = scratch[5] + scratch[4];
Fout[k+3*m] = scratch[5] - scratch[4];
}
}
inline
void bfly3( Complex * Fout, const size_t fstride, const size_t m)
{
size_t k=m;
const size_t m2 = 2*m;
Complex *tw1,*tw2;
Complex scratch[5];
Complex epi3;
epi3 = m_twiddles[fstride*m];
tw1=tw2=&m_twiddles[0];
do{
scratch[1]=Fout[m] * *tw1;
scratch[2]=Fout[m2] * *tw2;
scratch[3]=scratch[1]+scratch[2];
scratch[0]=scratch[1]-scratch[2];
tw1 += fstride;
tw2 += fstride*2;
Fout[m] = Complex( Fout->real() - .5*scratch[3].real() , Fout->imag() - .5*scratch[3].imag() );
scratch[0] *= epi3.imag();
*Fout += scratch[3];
Fout[m2] = Complex( Fout[m].real() + scratch[0].imag() , Fout[m].imag() - scratch[0].real() );
Fout[m] += Complex( -scratch[0].imag(),scratch[0].real() );
++Fout;
}while(--k);
}
inline
void bfly5( Complex * Fout, const size_t fstride, const size_t m)
{
Complex *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
size_t u;
Complex scratch[13];
Complex * twiddles = &m_twiddles[0];
Complex *tw;
Complex ya,yb;
ya = twiddles[fstride*m];
yb = twiddles[fstride*2*m];
Fout0=Fout;
Fout1=Fout0+m;
Fout2=Fout0+2*m;
Fout3=Fout0+3*m;
Fout4=Fout0+4*m;
tw=twiddles;
for ( u=0; u<m; ++u ) {
scratch[0] = *Fout0;
scratch[1] = *Fout1 * tw[u*fstride];
scratch[2] = *Fout2 * tw[2*u*fstride];
scratch[3] = *Fout3 * tw[3*u*fstride];
scratch[4] = *Fout4 * tw[4*u*fstride];
scratch[7] = scratch[1] + scratch[4];
scratch[10] = scratch[1] - scratch[4];
scratch[8] = scratch[2] + scratch[3];
scratch[9] = scratch[2] - scratch[3];
*Fout0 += scratch[7];
*Fout0 += scratch[8];
scratch[5] = scratch[0] + Complex(
(scratch[7].real()*ya.real() ) + (scratch[8].real() *yb.real() ),
(scratch[7].imag()*ya.real()) + (scratch[8].imag()*yb.real())
);
scratch[6] = Complex(
(scratch[10].imag()*ya.imag()) + (scratch[9].imag()*yb.imag()),
-(scratch[10].real()*ya.imag()) - (scratch[9].real()*yb.imag())
);
*Fout1 = scratch[5] - scratch[6];
*Fout4 = scratch[5] + scratch[6];
scratch[11] = scratch[0] +
Complex(
(scratch[7].real()*yb.real()) + (scratch[8].real()*ya.real()),
(scratch[7].imag()*yb.real()) + (scratch[8].imag()*ya.real())
);
scratch[12] = Complex(
-(scratch[10].imag()*yb.imag()) + (scratch[9].imag()*ya.imag()),
(scratch[10].real()*yb.imag()) - (scratch[9].real()*ya.imag())
);
*Fout2=scratch[11]+scratch[12];
*Fout3=scratch[11]-scratch[12];
++Fout0;++Fout1;++Fout2;++Fout3;++Fout4;
}
}
/* perform the butterfly for one stage of a mixed radix FFT */
inline
void bfly_generic(
Complex * Fout,
const size_t fstride,
int m,
int p
)
{
int u,k,q1,q;
Complex * twiddles = &m_twiddles[0];
Complex t;
int Norig = m_twiddles.size();
Complex * scratchbuf = &m_scratchBuf[0];
for ( u=0; u<m; ++u ) {
k=u;
for ( q1=0 ; q1<p ; ++q1 ) {
scratchbuf[q1] = Fout[ k ];
k += m;
}
k=u;
for ( q1=0 ; q1<p ; ++q1 ) {
int twidx=0;
Fout[ k ] = scratchbuf[0];
for (q=1;q<p;++q ) {
twidx += fstride * k;
if (twidx>=Norig) twidx-=Norig;
t=scratchbuf[q] * twiddles[twidx];
Fout[ k ] += t;
}
k += m;
}
}
}
};
template <typename _Scalar>
struct ei_kissfft_impl
{
typedef _Scalar Scalar;
typedef std::complex<Scalar> Complex;
void clear()
{
m_plans.clear();
m_realTwiddles.clear();
}
inline
void fwd( Complex * dst,const Complex *src,int nfft)
{
get_plan(nfft,false).work(0, dst, src, 1,1);
}
// real-to-complex forward FFT
// perform two FFTs of src even and src odd
// then twiddle to recombine them into the half-spectrum format
// then fill in the conjugate symmetric half
inline
void fwd( Complex * dst,const Scalar * src,int nfft)
{
if ( nfft&3 ) {
// use generic mode for odd
m_tmpBuf1.resize(nfft);
get_plan(nfft,false).work(0, &m_tmpBuf1[0], src, 1,1);
std::copy(m_tmpBuf1.begin(),m_tmpBuf1.begin()+(nfft>>1)+1,dst );
}else{
int ncfft = nfft>>1;
int ncfft2 = nfft>>2;
Complex * rtw = real_twiddles(ncfft2);
// use optimized mode for even real
fwd( dst, reinterpret_cast<const Complex*> (src), ncfft);
Complex dc = dst[0].real() + dst[0].imag();
Complex nyquist = dst[0].real() - dst[0].imag();
int k;
for ( k=1;k <= ncfft2 ; ++k ) {
Complex fpk = dst[k];
Complex fpnk = conj(dst[ncfft-k]);
Complex f1k = fpk + fpnk;
Complex f2k = fpk - fpnk;
Complex tw= f2k * rtw[k-1];
dst[k] = (f1k + tw) * Scalar(.5);
dst[ncfft-k] = conj(f1k -tw)*Scalar(.5);
}
dst[0] = dc;
dst[ncfft] = nyquist;
}
}
// inverse complex-to-complex
inline
void inv(Complex * dst,const Complex *src,int nfft)
{
get_plan(nfft,true).work(0, dst, src, 1,1);
}
// half-complex to scalar
inline
void inv( Scalar * dst,const Complex * src,int nfft)
{
if (nfft&3) {
m_tmpBuf1.resize(nfft);
m_tmpBuf2.resize(nfft);
std::copy(src,src+(nfft>>1)+1,m_tmpBuf1.begin() );
for (int k=1;k<(nfft>>1)+1;++k)
m_tmpBuf1[nfft-k] = conj(m_tmpBuf1[k]);
inv(&m_tmpBuf2[0],&m_tmpBuf1[0],nfft);
for (int k=0;k<nfft;++k)
dst[k] = m_tmpBuf2[k].real();
}else{
// optimized version for multiple of 4
int ncfft = nfft>>1;
int ncfft2 = nfft>>2;
Complex * rtw = real_twiddles(ncfft2);
m_tmpBuf1.resize(ncfft);
m_tmpBuf1[0] = Complex( src[0].real() + src[ncfft].real(), src[0].real() - src[ncfft].real() );
for (int k = 1; k <= ncfft / 2; ++k) {
Complex fk = src[k];
Complex fnkc = conj(src[ncfft-k]);
Complex fek = fk + fnkc;
Complex tmp = fk - fnkc;
Complex fok = tmp * conj(rtw[k-1]);
m_tmpBuf1[k] = fek + fok;
m_tmpBuf1[ncfft-k] = conj(fek - fok);
}
get_plan(ncfft,true).work(0, reinterpret_cast<Complex*>(dst), &m_tmpBuf1[0], 1,1);
}
}
protected:
typedef ei_kiss_cpx_fft<Scalar> PlanData;
typedef std::map<int,PlanData> PlanMap;
PlanMap m_plans;
std::map<int, std::vector<Complex> > m_realTwiddles;
std::vector<Complex> m_tmpBuf1;
std::vector<Complex> m_tmpBuf2;
inline
int PlanKey(int nfft,bool isinverse) const { return (nfft<<1) | isinverse; }
inline
PlanData & get_plan(int nfft,bool inverse)
{
// TODO look for PlanKey(nfft, ! inverse) and conjugate the twiddles
PlanData & pd = m_plans[ PlanKey(nfft,inverse) ];
if ( pd.m_twiddles.size() == 0 ) {
pd.make_twiddles(nfft,inverse);
pd.factorize(nfft);
}
return pd;
}
inline
Complex * real_twiddles(int ncfft2)
{
std::vector<Complex> & twidref = m_realTwiddles[ncfft2];// creates new if not there
if ( (int)twidref.size() != ncfft2 ) {
twidref.resize(ncfft2);
int ncfft= ncfft2<<1;
Scalar pi = acos( Scalar(-1) );
for (int k=1;k<=ncfft2;++k)
twidref[k-1] = exp( Complex(0,-pi * ((double) (k) / ncfft + .5) ) );
}
return &twidref[0];
}
};
/* vim: set filetype=cpp et sw=2 ts=2 ai: */