summaryrefslogblamecommitdiffstats
path: root/external/include/glm/gtx/scalar_multiplication.hpp
blob: b73edf676ab51f65e45c3d8225179f06b8c3a58d (plain) (tree)
1
2
3
4
5
6
7



                                           


                                                                                      










                                                                                                  



                                                                                                                                                                                                    











                                                                                                                  
                                          





                                                                                                             
                              
                                                   
                                        


                                                     
                              
                                                   
                                        


                                                     
                              
                                                   
                                        


















                                        
/// @ref gtx
/// @file glm/gtx/scalar_multiplication.hpp
/// @author Joshua Moerman
///
/// Include <glm/gtx/scalar_multiplication.hpp> to use the features of this extension.
///
/// Enables scalar multiplication for all types
///
/// Since GLSL is very strict about types, the following (often used) combinations do not work:
///    double * vec4
///    int * vec4
///    vec4 / int
/// So we'll fix that! Of course "float * vec4" should remain the same (hence the enable_if magic)

#pragma once

#include "../detail/setup.hpp"

#ifndef GLM_ENABLE_EXPERIMENTAL
#	error "GLM: GLM_GTX_scalar_multiplication is an experimental extension and may change in the future. Use #define GLM_ENABLE_EXPERIMENTAL before including it, if you really want to use it."
#endif

#if !GLM_HAS_TEMPLATE_ALIASES && !(GLM_COMPILER & GLM_COMPILER_GCC)
#	error "GLM_GTX_scalar_multiplication requires C++11 support or alias templates and if not support for GCC"
#endif

#include "../vec2.hpp"
#include "../vec3.hpp"
#include "../vec4.hpp"
#include "../mat2x2.hpp"
#include <type_traits>

namespace glm
{
	template<typename T, typename Vec>
	using return_type_scalar_multiplication = typename std::enable_if<
		!std::is_same<T, float>::value       // T may not be a float
		&& std::is_arithmetic<T>::value, Vec // But it may be an int or double (no vec3 or mat3, ...)
	>::type;

#define GLM_IMPLEMENT_SCAL_MULT(Vec) \
	template<typename T> \
	return_type_scalar_multiplication<T, Vec> \
	operator*(T const& s, Vec rh){ \
		return rh *= static_cast<float>(s); \
	} \
	 \
	template<typename T> \
	return_type_scalar_multiplication<T, Vec> \
	operator*(Vec lh, T const& s){ \
		return lh *= static_cast<float>(s); \
	} \
	 \
	template<typename T> \
	return_type_scalar_multiplication<T, Vec> \
	operator/(Vec lh, T const& s){ \
		return lh *= 1.0f / s; \
	}

GLM_IMPLEMENT_SCAL_MULT(vec2)
GLM_IMPLEMENT_SCAL_MULT(vec3)
GLM_IMPLEMENT_SCAL_MULT(vec4)

GLM_IMPLEMENT_SCAL_MULT(mat2)
GLM_IMPLEMENT_SCAL_MULT(mat2x3)
GLM_IMPLEMENT_SCAL_MULT(mat2x4)
GLM_IMPLEMENT_SCAL_MULT(mat3x2)
GLM_IMPLEMENT_SCAL_MULT(mat3)
GLM_IMPLEMENT_SCAL_MULT(mat3x4)
GLM_IMPLEMENT_SCAL_MULT(mat4x2)
GLM_IMPLEMENT_SCAL_MULT(mat4x3)
GLM_IMPLEMENT_SCAL_MULT(mat4)

#undef GLM_IMPLEMENT_SCAL_MULT
} // namespace glm