Projekt-GRK/dependencies/glm/detail/func_exponential.hpp

104 lines
5.4 KiB
C++
Raw Normal View History

2018-06-20 21:34:24 +02:00
/// @ref core
/// @file glm/detail/func_exponential.hpp
///
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
///
/// @defgroup core_func_exponential Exponential functions
/// @ingroup core
///
/// These all operate component-wise. The description is per component.
#pragma once
#include "type_vec1.hpp"
#include "type_vec2.hpp"
#include "type_vec3.hpp"
#include "type_vec4.hpp"
#include <cmath>
namespace glm
{
/// @addtogroup core_func_exponential
/// @{
/// Returns 'base' raised to the power 'exponent'.
///
/// @param base Floating point value. pow function is defined for input values of 'base' defined in the range (inf-, inf+) in the limit of the type precision.
/// @param exponent Floating point value representing the 'exponent'.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/pow.xml">GLSL pow man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> pow(vecType<T, P> const & base, vecType<T, P> const & exponent);
/// Returns the natural exponentiation of x, i.e., e^x.
///
/// @param v exp function is defined for input values of v defined in the range (inf-, inf+) in the limit of the type precision.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/exp.xml">GLSL exp man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> exp(vecType<T, P> const & v);
/// Returns the natural logarithm of v, i.e.,
/// returns the value y which satisfies the equation x = e^y.
/// Results are undefined if v <= 0.
///
/// @param v log function is defined for input values of v defined in the range (0, inf+) in the limit of the type precision.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/log.xml">GLSL log man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> log(vecType<T, P> const & v);
/// Returns 2 raised to the v power.
///
/// @param v exp2 function is defined for input values of v defined in the range (inf-, inf+) in the limit of the type precision.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/exp2.xml">GLSL exp2 man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> exp2(vecType<T, P> const & v);
/// Returns the base 2 log of x, i.e., returns the value y,
/// which satisfies the equation x = 2 ^ y.
///
/// @param v log2 function is defined for input values of v defined in the range (0, inf+) in the limit of the type precision.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/log2.xml">GLSL log2 man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> log2(vecType<T, P> const & v);
/// Returns the positive square root of v.
///
/// @param v sqrt function is defined for input values of v defined in the range [0, inf+) in the limit of the type precision.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/sqrt.xml">GLSL sqrt man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
//template <typename genType>
//GLM_FUNC_DECL genType sqrt(genType const & x);
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> sqrt(vecType<T, P> const & v);
/// Returns the reciprocal of the positive square root of v.
///
/// @param v inversesqrt function is defined for input values of v defined in the range [0, inf+) in the limit of the type precision.
/// @tparam genType Floating-point scalar or vector types.
///
/// @see <a href="http://www.opengl.org/sdk/docs/manglsl/xhtml/inversesqrt.xml">GLSL inversesqrt man page</a>
/// @see <a href="http://www.opengl.org/registry/doc/GLSLangSpec.4.20.8.pdf">GLSL 4.20.8 specification, section 8.2 Exponential Functions</a>
template <typename T, precision P, template <typename, precision> class vecType>
GLM_FUNC_DECL vecType<T, P> inversesqrt(vecType<T, P> const & v);
/// @}
}//namespace glm
#include "func_exponential.inl"