kicad/include/gal/opengl/glm/gtc/half_float.hpp

441 lines
14 KiB
C++
Raw Normal View History

2013-07-16 06:42:44 +00:00
///////////////////////////////////////////////////////////////////////////////////
/// OpenGL Mathematics (glm.g-truc.net)
///
/// Copyright (c) 2005 - 2012 G-Truc Creation (www.g-truc.net)
/// Permission is hereby granted, free of charge, to any person obtaining a copy
/// of this software and associated documentation files (the "Software"), to deal
/// in the Software without restriction, including without limitation the rights
/// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
/// copies of the Software, and to permit persons to whom the Software is
/// furnished to do so, subject to the following conditions:
///
/// The above copyright notice and this permission notice shall be included in
/// all copies or substantial portions of the Software.
///
/// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
/// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
/// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
/// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
/// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
/// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
/// THE SOFTWARE.
///
/// @ref gtc_half_float
/// @file glm/gtc/half_float.hpp
/// @date 2009-04-29 / 2012-11-06
/// @author Christophe Riccio
///
/// @see core (dependence)
///
/// @defgroup gtc_half_float GLM_GTC_half_float
/// @ingroup gtc
///
/// Defines the half-precision floating-point type, along with various typedefs for vectors and matrices.
/// <glm/gtc/half_float.hpp> need to be included to use these functionalities.
///////////////////////////////////////////////////////////////////////////////////
#ifndef GLM_GTC_half_float
#define GLM_GTC_half_float GLM_VERSION
// Dependency:
#include "../glm.hpp"
#if(defined(GLM_MESSAGES) && !defined(glm_ext))
# pragma message("GLM: GLM_GTC_half_float extension included")
#endif
namespace glm{
namespace detail
{
#if(GLM_COMPONENT == GLM_COMPONENT_CXX98)
template <>
struct tvec2<half>
{
enum ctor{null};
typedef half value_type;
typedef std::size_t size_type;
GLM_FUNC_DECL size_type length() const;
static GLM_FUNC_DECL size_type value_size();
typedef tvec2<half> type;
typedef tvec2<bool> bool_type;
//////////////////////////////////////
// Data
half x, y;
//////////////////////////////////////
// Accesses
half & operator[](size_type i);
half const & operator[](size_type i) const;
//////////////////////////////////////
// Implicit basic constructors
tvec2();
tvec2(tvec2<half> const & v);
//////////////////////////////////////
// Explicit basic constructors
explicit tvec2(ctor);
explicit tvec2(
half const & s);
explicit tvec2(
half const & s1,
half const & s2);
//////////////////////////////////////
// Swizzle constructors
tvec2(tref2<half> const & r);
//////////////////////////////////////
// Convertion scalar constructors
//! Explicit converions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec2(U const & x);
//! Explicit converions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U, typename V>
explicit tvec2(U const & x, V const & y);
//////////////////////////////////////
// Convertion vector constructors
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec2(tvec2<U> const & v);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec2(tvec3<U> const & v);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec2(tvec4<U> const & v);
//////////////////////////////////////
// Unary arithmetic operators
tvec2<half>& operator= (tvec2<half> const & v);
tvec2<half>& operator+=(half const & s);
tvec2<half>& operator+=(tvec2<half> const & v);
tvec2<half>& operator-=(half const & s);
tvec2<half>& operator-=(tvec2<half> const & v);
tvec2<half>& operator*=(half const & s);
tvec2<half>& operator*=(tvec2<half> const & v);
tvec2<half>& operator/=(half const & s);
tvec2<half>& operator/=(tvec2<half> const & v);
tvec2<half>& operator++();
tvec2<half>& operator--();
//////////////////////////////////////
// Swizzle operators
half swizzle(comp X) const;
tvec2<half> swizzle(comp X, comp Y) const;
tvec3<half> swizzle(comp X, comp Y, comp Z) const;
tvec4<half> swizzle(comp X, comp Y, comp Z, comp W) const;
tref2<half> swizzle(comp X, comp Y);
};
template <>
struct tvec3<half>
{
enum ctor{null};
typedef half value_type;
typedef std::size_t size_type;
GLM_FUNC_DECL size_type length() const;
static GLM_FUNC_DECL size_type value_size();
typedef tvec3<half> type;
typedef tvec3<bool> bool_type;
//////////////////////////////////////
// Data
half x, y, z;
//////////////////////////////////////
// Accesses
half & operator[](size_type i);
half const & operator[](size_type i) const;
//////////////////////////////////////
// Implicit basic constructors
tvec3();
tvec3(tvec3<half> const & v);
//////////////////////////////////////
// Explicit basic constructors
explicit tvec3(ctor);
explicit tvec3(
half const & s);
explicit tvec3(
half const & s1,
half const & s2,
half const & s3);
//////////////////////////////////////
// Swizzle constructors
tvec3(tref3<half> const & r);
//////////////////////////////////////
// Convertion scalar constructors
//! Explicit converions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec3(U const & x);
//! Explicit converions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U, typename V, typename W>
explicit tvec3(U const & x, V const & y, W const & z);
//////////////////////////////////////
// Convertion vector constructors
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B>
explicit tvec3(tvec2<A> const & v, B const & s);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B>
explicit tvec3(A const & s, tvec2<B> const & v);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec3(tvec3<U> const & v);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec3(tvec4<U> const & v);
//////////////////////////////////////
// Unary arithmetic operators
tvec3<half>& operator= (tvec3<half> const & v);
tvec3<half>& operator+=(half const & s);
tvec3<half>& operator+=(tvec3<half> const & v);
tvec3<half>& operator-=(half const & s);
tvec3<half>& operator-=(tvec3<half> const & v);
tvec3<half>& operator*=(half const & s);
tvec3<half>& operator*=(tvec3<half> const & v);
tvec3<half>& operator/=(half const & s);
tvec3<half>& operator/=(tvec3<half> const & v);
tvec3<half>& operator++();
tvec3<half>& operator--();
//////////////////////////////////////
// Swizzle operators
half swizzle(comp X) const;
tvec2<half> swizzle(comp X, comp Y) const;
tvec3<half> swizzle(comp X, comp Y, comp Z) const;
tvec4<half> swizzle(comp X, comp Y, comp Z, comp W) const;
tref3<half> swizzle(comp X, comp Y, comp Z);
};
template <>
struct tvec4<half>
{
enum ctor{null};
typedef half value_type;
typedef std::size_t size_type;
GLM_FUNC_DECL size_type length() const;
static GLM_FUNC_DECL size_type value_size();
typedef tvec4<half> type;
typedef tvec4<bool> bool_type;
//////////////////////////////////////
// Data
half x, y, z, w;
//////////////////////////////////////
// Accesses
half & operator[](size_type i);
half const & operator[](size_type i) const;
//////////////////////////////////////
// Implicit basic constructors
tvec4();
tvec4(tvec4<half> const & v);
//////////////////////////////////////
// Explicit basic constructors
explicit tvec4(ctor);
explicit tvec4(
half const & s);
explicit tvec4(
half const & s0,
half const & s1,
half const & s2,
half const & s3);
//////////////////////////////////////
// Swizzle constructors
tvec4(tref4<half> const & r);
//////////////////////////////////////
// Convertion scalar constructors
//! Explicit converions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec4(U const & x);
//! Explicit converions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B, typename C, typename D>
explicit tvec4(A const & x, B const & y, C const & z, D const & w);
//////////////////////////////////////
// Convertion vector constructors
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B, typename C>
explicit tvec4(tvec2<A> const & v, B const & s1, C const & s2);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B, typename C>
explicit tvec4(A const & s1, tvec2<B> const & v, C const & s2);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B, typename C>
explicit tvec4(A const & s1, B const & s2, tvec2<C> const & v);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B>
explicit tvec4(tvec3<A> const & v, B const & s);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B>
explicit tvec4(A const & s, tvec3<B> const & v);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename A, typename B>
explicit tvec4(tvec2<A> const & v1, tvec2<B> const & v2);
//! Explicit conversions (From section 5.4.1 Conversion and scalar constructors of GLSL 1.30.08 specification)
template <typename U>
explicit tvec4(tvec4<U> const & v);
//////////////////////////////////////
// Unary arithmetic operators
tvec4<half>& operator= (tvec4<half> const & v);
tvec4<half>& operator+=(half const & s);
tvec4<half>& operator+=(tvec4<half> const & v);
tvec4<half>& operator-=(half const & s);
tvec4<half>& operator-=(tvec4<half> const & v);
tvec4<half>& operator*=(half const & s);
tvec4<half>& operator*=(tvec4<half> const & v);
tvec4<half>& operator/=(half const & s);
tvec4<half>& operator/=(tvec4<half> const & v);
tvec4<half>& operator++();
tvec4<half>& operator--();
//////////////////////////////////////
// Swizzle operators
half swizzle(comp X) const;
tvec2<half> swizzle(comp X, comp Y) const;
tvec3<half> swizzle(comp X, comp Y, comp Z) const;
tvec4<half> swizzle(comp X, comp Y, comp Z, comp W) const;
tref4<half> swizzle(comp X, comp Y, comp Z, comp W);
};
#endif//(GLM_COMPONENT == GLM_COMPONENT_CXX98)
}
//namespace detail
/// @addtogroup gtc_half_float
/// @{
/// Type for half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::half half;
/// Vector of 2 half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tvec2<detail::half> hvec2;
/// Vector of 3 half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tvec3<detail::half> hvec3;
/// Vector of 4 half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tvec4<detail::half> hvec4;
/// 2 * 2 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat2x2<detail::half> hmat2;
/// 3 * 3 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat3x3<detail::half> hmat3;
/// 4 * 4 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat4x4<detail::half> hmat4;
/// 2 * 2 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat2x2<detail::half> hmat2x2;
/// 2 * 3 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat2x3<detail::half> hmat2x3;
/// 2 * 4 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat2x4<detail::half> hmat2x4;
/// 3 * 2 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat3x2<detail::half> hmat3x2;
/// 3 * 3 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat3x3<detail::half> hmat3x3;
/// 3 * 4 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat3x4<detail::half> hmat3x4;
/// 4 * 2 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat4x2<detail::half> hmat4x2;
/// 4 * 3 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat4x3<detail::half> hmat4x3;
/// 4 * 4 matrix of half-precision floating-point numbers.
/// @see gtc_half_float
typedef detail::tmat4x4<detail::half> hmat4x4;
/// Returns the absolute value of a half-precision floating-point value
/// @see gtc_half_float
half abs(half const & x);
/// Returns the absolute value of a half-precision floating-point two dimensional vector
/// @see gtc_half_float
hvec2 abs(hvec2 const & x);
/// Returns the absolute value of a half-precision floating-point three dimensional vector
/// @see gtc_half_float
hvec3 abs(hvec3 const & x);
/// Returns the absolute value of a half-precision floating-point four dimensional vector
/// @see gtc_half_float
hvec4 abs(hvec4 const & x);
/// @}
}// namespace glm
#include "half_float.inl"
#endif//GLM_GTC_half_float