You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

470 lines
13 KiB

// David Eberly, Geometric Tools, Redmond WA 98052
// Copyright (c) 1998-2021
// Distributed under the Boost Software License, Version 1.0.
// https://www.boost.org/LICENSE_1_0.txt
// https://www.geometrictools.com/License/Boost/LICENSE_1_0.txt
// Version: 5.5.2021.01.14
#pragma once
#include <Mathematics/Logger.h>
#include <Mathematics/Math.h>
#include <array>
// The SWInterval [e0,e1] must satisfy e0 <= e1. Expose this define to trap
// invalid construction where e0 > e1.
#define GTE_THROW_ON_INVALID_SWINTERVAL
namespace gte
{
// The T must be 'float' or 'double'.
template <typename T>
class SWInterval
{
public:
// Convenient constants.
static T constexpr zero = 0;
static T constexpr one = 1;
static T constexpr max = std::numeric_limits<T>::max();
static T constexpr inf = std::numeric_limits<T>::infinity();
// Construction. This is the only way to create an interval. All such
// intervals are immutable once created. The constructor SWInterval(T)
// is used to create the degenerate interval [e,e].
SWInterval()
:
mEndpoints{ static_cast<T>(0), static_cast<T>(0) }
{
static_assert(std::is_floating_point<T>::value, "Invalid type.");
}
SWInterval(SWInterval const& other)
:
mEndpoints(other.mEndpoints)
{
static_assert(std::is_floating_point<T>::value, "Invalid type.");
}
SWInterval(T e)
:
mEndpoints{ e, e }
{
static_assert(std::is_floating_point<T>::value, "Invalid type.");
}
SWInterval(T e0, T e1)
:
mEndpoints{ e0, e1 }
{
static_assert(std::is_floating_point<T>::value, "Invalid type.");
#if defined(GTE_THROW_ON_INVALID_SWINTERVAL)
LogAssert(mEndpoints[0] <= mEndpoints[1], "Invalid SWInterval.");
#endif
}
SWInterval(std::array<T, 2> const& endpoint)
:
mEndpoints(endpoint)
{
static_assert(std::is_floating_point<T>::value, "Invalid type.");
#if defined(GTE_THROW_ON_INVALID_SWINTERVAL)
LogAssert(mEndpoints[0] <= mEndpoints[1], "Invalid SWInterval.");
#endif
}
SWInterval& operator=(SWInterval const& other)
{
static_assert(std::is_floating_point<T>::value, "Invalid type.");
mEndpoints = other.mEndpoints;
return *this;
}
// Member access. It is only possible to read the endpoints. You
// cannot modify the endpoints outside the arithmetic operations.
inline T operator[](size_t i) const
{
return mEndpoints[i];
}
inline std::array<T, 2> GetEndpoints() const
{
return mEndpoints;
}
// Arithmetic operations to compute intervals at the leaf nodes of
// an expression tree. Such nodes correspond to the raw floating-point
// variables of the expression. The non-class operators defined after
// the class definition are used to compute intervals at the interior
// nodes of the expression tree.
inline static SWInterval Add(T u, T v)
{
SWInterval w;
T add = u + v;
w.mEndpoints[0] = std::nextafter(add, -max);
w.mEndpoints[1] = std::nextafter(add, +max);
return w;
}
inline static SWInterval Sub(T u, T v)
{
SWInterval w;
T sub = u - v;
w.mEndpoints[0] = std::nextafter(sub, -max);
w.mEndpoints[1] = std::nextafter(sub, +max);
return w;
}
inline static SWInterval Mul(T u, T v)
{
SWInterval w;
T mul = u * v;
w.mEndpoints[0] = std::nextafter(mul, -max);
w.mEndpoints[1] = std::nextafter(mul, +max);
return w;
}
inline static SWInterval Div(T u, T v)
{
if (v != zero)
{
SWInterval w;
T div = u / v;
w.mEndpoints[0] = std::nextafter(div, -max);
w.mEndpoints[1] = std::nextafter(div, +max);
return w;
}
else
{
// Division by zero does not lead to a determinate SWInterval.
// Return the entire set of real numbers.
return Reals();
}
}
private:
std::array<T, 2> mEndpoints;
public:
// FOR INTERNAL USE ONLY. These are used by the non-class operators
// defined after the class definition.
inline static SWInterval Add(T u0, T u1, T v0, T v1)
{
SWInterval w;
w.mEndpoints[0] = std::nextafter(u0 + v0, -max);
w.mEndpoints[1] = std::nextafter(u1 + v1, +max);
return w;
}
inline static SWInterval Sub(T u0, T u1, T v0, T v1)
{
SWInterval w;
w.mEndpoints[0] = std::nextafter(u0 - v1, -max);
w.mEndpoints[1] = std::nextafter(u1 - v0, +max);
return w;
}
inline static SWInterval Mul(T u0, T u1, T v0, T v1)
{
SWInterval w;
w.mEndpoints[0] = std::nextafter(u0 * v0, -max);
w.mEndpoints[1] = std::nextafter(u1 * v1, +max);
return w;
}
inline static SWInterval Mul2(T u0, T u1, T v0, T v1)
{
T u0mv1 = std::nextafter(u0 * v1, -max);
T u1mv0 = std::nextafter(u1 * v0, -max);
T u0mv0 = std::nextafter(u0 * v0, +max);
T u1mv1 = std::nextafter(u1 * v1, +max);
return SWInterval<T>(std::min(u0mv1, u1mv0), std::max(u0mv0, u1mv1));
}
inline static SWInterval Div(T u0, T u1, T v0, T v1)
{
SWInterval w;
w.mEndpoints[0] = std::nextafter(u0 / v1, -max);
w.mEndpoints[1] = std::nextafter(u1 / v0, +max);
return w;
}
inline static SWInterval Reciprocal(T v0, T v1)
{
SWInterval w;
w.mEndpoints[0] = std::nextafter(one / v1, -max);
w.mEndpoints[1] = std::nextafter(one / v0, +max);
return w;
}
inline static SWInterval ReciprocalDown(T v)
{
T recpv = std::nextafter(one / v, -max);
return SWInterval<T>(recpv, +inf);
}
inline static SWInterval ReciprocalUp(T v)
{
T recpv = std::nextafter(one / v, +max);
return SWInterval<T>(-inf, recpv);
}
inline static SWInterval Reals()
{
return SWInterval(-inf, +inf);
}
};
// Unary operations. Negation of [e0,e1] produces [-e1,-e0]. This
// operation needs to be supported in the sense of negating a
// "number" in an arithmetic expression.
template <typename T>
SWInterval<T> operator+(SWInterval<T> const& u)
{
return u;
}
template <typename T>
SWInterval<T> operator-(SWInterval<T> const& u)
{
return SWInterval<T>(-u[1], -u[0]);
}
// Addition operations.
template <typename T>
SWInterval<T> operator+(T u, SWInterval<T> const& v)
{
return SWInterval<T>::Add(u, u, v[0], v[1]);
}
template <typename T>
SWInterval<T> operator+(SWInterval<T> const& u, T v)
{
return SWInterval<T>::Add(u[0], u[1], v, v);
}
template <typename T>
SWInterval<T> operator+(SWInterval<T> const& u, SWInterval<T> const& v)
{
return SWInterval<T>::Add(u[0], u[1], v[0], v[1]);
}
template <typename T>
SWInterval<T>& operator+=(SWInterval<T>& u, T v)
{
u = u + v;
return u;
}
template <typename T>
SWInterval<T>& operator+=(SWInterval<T>& u, SWInterval<T> const& v)
{
u = u + v;
return u;
}
// Subtraction operations.
template <typename T>
SWInterval<T> operator-(T u, SWInterval<T> const& v)
{
return SWInterval<T>::Sub(u, u, v[0], v[1]);
}
template <typename T>
SWInterval<T> operator-(SWInterval<T> const& u, T v)
{
return SWInterval<T>::Sub(u[0], u[1], v, v);
}
template <typename T>
SWInterval<T> operator-(SWInterval<T> const& u, SWInterval<T> const& v)
{
return SWInterval<T>::Sub(u[0], u[1], v[0], v[1]);
}
template <typename T>
SWInterval<T>& operator-=(SWInterval<T>& u, T v)
{
u = u - v;
return u;
}
template <typename T>
SWInterval<T>& operator-=(SWInterval<T>& u, SWInterval<T> const& v)
{
u = u - v;
return u;
}
// Multiplication operations.
template <typename T>
SWInterval<T> operator*(T u, SWInterval<T> const& v)
{
if (u >= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u, u, v[0], v[1]);
}
else
{
return SWInterval<T>::Mul(u, u, v[1], v[0]);
}
}
template <typename T>
SWInterval<T> operator*(SWInterval<T> const& u, T v)
{
if (v >= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[0], u[1], v, v);
}
else
{
return SWInterval<T>::Mul(u[1], u[0], v, v);
}
}
template <typename T>
SWInterval<T> operator*(SWInterval<T> const& u, SWInterval<T> const& v)
{
if (u[0] >= SWInterval<T>::zero)
{
if (v[0] >= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[0], u[1], v[0], v[1]);
}
else if (v[1] <= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[1], u[0], v[0], v[1]);
}
else // v[0] < 0 < v[1]
{
return SWInterval<T>::Mul(u[1], u[1], v[0], v[1]);
}
}
else if (u[1] <= SWInterval<T>::zero)
{
if (v[0] >= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[0], u[1], v[1], v[0]);
}
else if (v[1] <= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[1], u[0], v[1], v[0]);
}
else // v[0] < 0 < v[1]
{
return SWInterval<T>::Mul(u[0], u[0], v[1], v[0]);
}
}
else // u[0] < 0 < u[1]
{
if (v[0] >= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[0], u[1], v[1], v[1]);
}
else if (v[1] <= SWInterval<T>::zero)
{
return SWInterval<T>::Mul(u[1], u[0], v[0], v[0]);
}
else // v[0] < 0 < v[1]
{
return SWInterval<T>::Mul2(u[0], u[1], v[0], v[1]);
}
}
}
template <typename T>
SWInterval<T>& operator*=(SWInterval<T>& u, T v)
{
u = u * v;
return u;
}
template <typename T>
SWInterval<T>& operator*=(SWInterval<T>& u, SWInterval<T> const& v)
{
u = u * v;
return u;
}
// Division operations. If the divisor SWInterval is [v0,v1] with
// v0 < 0 < v1, then the returned SWInterval is (-inf,+inf) instead of
// Union((-inf,1/v0),(1/v1,+inf)). An application should try to avoid
// this case by branching based on [v0,0] and [0,v1].
template <typename T>
SWInterval<T> operator/(T u, SWInterval<T> const& v)
{
if (v[0] > SWInterval<T>::zero || v[1] < SWInterval<T>::zero)
{
return u * SWInterval<T>::Reciprocal(v[0], v[1]);
}
else
{
if (v[0] == SWInterval<T>::zero)
{
return u * SWInterval<T>::ReciprocalDown(v[1]);
}
else if (v[1] == SWInterval<T>::zero)
{
return u * SWInterval<T>::ReciprocalUp(v[0]);
}
else // v[0] < 0 < v[1]
{
return SWInterval<T>::Reals();
}
}
}
template <typename T>
SWInterval<T> operator/(SWInterval<T> const& u, T v)
{
if (v > SWInterval<T>::zero)
{
return SWInterval<T>::Div(u[0], u[1], v, v);
}
else if (v < SWInterval<T>::zero)
{
return SWInterval<T>::Div(u[1], u[0], v, v);
}
else // v = 0
{
return SWInterval<T>::Reals();
}
}
template <typename T>
SWInterval<T> operator/(SWInterval<T> const& u, SWInterval<T> const& v)
{
if (v[0] > SWInterval<T>::zero || v[1] < SWInterval<T>::zero)
{
return u * SWInterval<T>::Reciprocal(v[0], v[1]);
}
else
{
if (v[0] == SWInterval<T>::zero)
{
return u * SWInterval<T>::ReciprocalDown(v[1]);
}
else if (v[1] == SWInterval<T>::zero)
{
return u * SWInterval<T>::ReciprocalUp(v[0]);
}
else // v[0] < 0 < v[1]
{
return SWInterval<T>::Reals();
}
}
}
template <typename T>
SWInterval<T>& operator/=(SWInterval<T>& u, T v)
{
u = u / v;
return u;
}
template <typename T>
SWInterval<T>& operator/=(SWInterval<T>& u, SWInterval<T> const& v)
{
u = u / v;
return u;
}
}