mirror of
https://github.com/g-truc/glm.git
synced 2024-11-17 15:24:35 +00:00
175 lines
5.7 KiB
C++
175 lines
5.7 KiB
C++
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
||
|
// OpenGL Mathematics Copyright (c) 2005 - 2008 G-Truc Creation (www.g-truc.net)
|
||
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
||
|
// Created : 2005-01-14
|
||
|
// Updated : 2009-05-01
|
||
|
// Licence : This source is under MIT License
|
||
|
// File : glm/glm.hpp
|
||
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
||
|
|
||
|
/*! \mainpage OpenGL Mathematics
|
||
|
*
|
||
|
* OpenGL Mathematics (GLM) is a C++ mathematics library for 3D applications based on the OpenGL Shading Language (GLSL) specification.
|
||
|
*
|
||
|
* The goal of the project is to provide to 3D programmers math classes and functions that miss in C++ when we use to program with GLSL or any high level GPU language. With GLM, the idea is to have a library that works the same way that GLSL which imply a strict following of GLSL specification for the implementation.
|
||
|
*
|
||
|
* However, this project isn't limited by GLSL features. An extension system based on GLSL extensions development conventions allows to extend GLSL capabilities.
|
||
|
*
|
||
|
* GLM is release under MIT license and available for all version of GCC from version 3.4 and Visual Studio from version 8.0 as a platform independent library.
|
||
|
*
|
||
|
* Any feedback is welcome, please send them to g.truc.creation[NO_SPAM_THANKS]gmail.com.
|
||
|
*
|
||
|
*/
|
||
|
|
||
|
#ifndef glm_glm
|
||
|
#define glm_glm
|
||
|
|
||
|
#ifdef max
|
||
|
#undef max
|
||
|
#endif
|
||
|
|
||
|
#ifdef min
|
||
|
#undef min
|
||
|
#endif
|
||
|
|
||
|
#define GLMvalType typename genType::value_type
|
||
|
#define GLMcolType typename genType::col_type
|
||
|
#define GLMrowType typename genType::row_type
|
||
|
|
||
|
#define GLMsizeType typename genType::size_type
|
||
|
#define GLMrowSize typename genType::row_size
|
||
|
#define GLMcolSize typename genType::col_size
|
||
|
|
||
|
#include <cmath>
|
||
|
#include <climits>
|
||
|
#include <limits>
|
||
|
#include "./setup.hpp"
|
||
|
|
||
|
//! GLM namespace, it contains all GLSL based features.
|
||
|
namespace glm
|
||
|
{
|
||
|
namespace test
|
||
|
{
|
||
|
bool main_bug();
|
||
|
bool main_core();
|
||
|
}//namespace test
|
||
|
|
||
|
//! GLM core. Namespace that includes all the feature define by GLSL 1.30.8 specification. This namespace is included in glm namespace.
|
||
|
namespace core
|
||
|
{
|
||
|
//! Scalar, vectors and matrices
|
||
|
//! from section 4.1.2 Booleans, 4.1.3 Integers section, 4.1.4 Floats section,
|
||
|
//! 4.1.5 Vectors and section 4.1.6 Matrices of GLSL 1.30.8 specification.
|
||
|
//! This namespace resolves precision qualifier define in section 4.5 of GLSL 1.30.8 specification.
|
||
|
namespace type
|
||
|
{
|
||
|
//! Scalar types from section 4.1.2 Booleans, 4.1.3 Integers and 4.1.4 Floats of GLSL 1.30.8 specification.
|
||
|
//! This namespace is included in glm namespace.
|
||
|
namespace scalar
|
||
|
{
|
||
|
//! Scalar types with precision qualifier.
|
||
|
//! This namespace is included in glm namespace.
|
||
|
namespace precision{}
|
||
|
}
|
||
|
|
||
|
//! Vector types from section 4.1.5 of GLSL 1.30.8 specification.
|
||
|
//! This namespace is included in glm namespace.
|
||
|
namespace vector
|
||
|
{
|
||
|
//! Vector types with precision qualifier.
|
||
|
//! This namespace is included in glm namespace.
|
||
|
namespace precision{}
|
||
|
}
|
||
|
|
||
|
//! Matrix types from section 4.1.6 of GLSL 1.30.8 specification.
|
||
|
//! This namespace is included in glm namespace.
|
||
|
namespace matrix
|
||
|
{
|
||
|
//! Matrix types with precision qualifier.
|
||
|
//! This namespace is included in glm namespace.
|
||
|
namespace precision{}
|
||
|
}
|
||
|
}
|
||
|
//! Some of the functions defined in section 8 Built-in Functions of GLSL 1.30.8 specification.
|
||
|
//! Angle and trigonometry, exponential, common, geometric, matrix and vector relational functions.
|
||
|
namespace function{}
|
||
|
}
|
||
|
//namespace core
|
||
|
|
||
|
using namespace core::type::scalar;
|
||
|
using namespace core::type::scalar::precision;
|
||
|
using namespace core::type::vector;
|
||
|
using namespace core::type::vector::precision;
|
||
|
using namespace core::type::matrix;
|
||
|
using namespace core::type::matrix::precision;
|
||
|
|
||
|
//! GLM experimental extensions. The interface could change between releases.
|
||
|
namespace gtx{}
|
||
|
|
||
|
//! GLM stable extensions.
|
||
|
namespace gtc{}
|
||
|
|
||
|
//! IMG extensions.
|
||
|
namespace img{}
|
||
|
|
||
|
//! VIRTREV extensions.
|
||
|
namespace img{}
|
||
|
|
||
|
} //namespace glm
|
||
|
|
||
|
#include "./core/_detail.hpp"
|
||
|
#include "./core/type.hpp"
|
||
|
#include "./core/type_half.hpp"
|
||
|
|
||
|
#include "./core/func_common.hpp"
|
||
|
#include "./core/func_exponential.hpp"
|
||
|
#include "./core/func_geometric.hpp"
|
||
|
#include "./core/func_matrix.hpp"
|
||
|
#include "./core/func_trigonometric.hpp"
|
||
|
#include "./core/func_vector_relational.hpp"
|
||
|
#include "./core/func_noise.hpp"
|
||
|
#include "./core/_swizzle.hpp"
|
||
|
//#include "./core/_xref2.hpp"
|
||
|
//#include "./core/_xref3.hpp"
|
||
|
//#include "./core/_xref4.hpp"
|
||
|
|
||
|
#if(defined(GLM_MESSAGE) && (GLM_MESSAGE & (GLM_MESSAGE_CORE | GLM_MESSAGE_NOTIFICATION)))
|
||
|
# pragma message("GLM message: Core library included")
|
||
|
#endif//GLM_MESSAGE
|
||
|
|
||
|
#if(defined(GLM_COMPILER) && (GLM_COMPILER & GLM_COMPILER_VC))
|
||
|
|
||
|
#define GLM_DEPRECATED __declspec(deprecated)
|
||
|
#define GLM_RESTRICT __restrict
|
||
|
#define GLM_ALIGN(x) __declspec(align(x))
|
||
|
|
||
|
//#define aligned(x) __declspec(align(x)) struct
|
||
|
|
||
|
#else
|
||
|
|
||
|
#define GLM_DEPRECATED
|
||
|
#define GLM_RESTRICT
|
||
|
#define GLM_ALIGN(x)
|
||
|
|
||
|
#endif//GLM_COMPILER
|
||
|
|
||
|
////////////////////
|
||
|
// check type sizes
|
||
|
#ifndef GLM_STATIC_ASSERT_NULL
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::int8)==1);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::int16)==2);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::int32)==4);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::int64)==8);
|
||
|
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::uint8)==1);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::uint16)==2);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::uint32)==4);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::uint64)==8);
|
||
|
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::float16)==2);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::float32)==4);
|
||
|
GLM_STATIC_ASSERT(sizeof(glm::detail::float64)==8);
|
||
|
#endif//GLM_STATIC_ASSERT_NULL
|
||
|
|
||
|
#endif //glm_glm
|