Fixed build

This commit is contained in:
Christophe Riccio
2011-05-05 13:36:07 +01:00
parent 378da68899
commit 06d0b33f7b
2 changed files with 32 additions and 108 deletions

View File

@@ -7,12 +7,19 @@
// File : glm/gtx/ulp.inl
///////////////////////////////////////////////////////////////////////////////////////////////////
#include <cmath>
#include <cfloat>
#if(GLM_COMPILER & GLM_COMPILER_VC)
# include <cfloat>
# define GLM_NEXT_AFTER _nextafterf
# if(GLM_MODEL == GLM_MODEL_32)
# define GLM_NEXT_AFTER_FLT(x, toward) (float)_nextafter(x, float(toward))
# else
# define GLM_NEXT_AFTER_FLT(x, toward) _nextafterf(x, toward)
# endif
# define GLM_NEXT_AFTER_DBL(x, toward) _nextafter(x, toward)
#else
# include <cmath>
# define GLM_NEXT_AFTER nextafterf
# define GLM_NEXT_AFTER_FLT(x, toward) nextafterf(x, toward)
# define GLM_NEXT_AFTER_DBL(x, toward) nextafter(x, toward)
#endif
namespace glm{
@@ -21,12 +28,12 @@ namespace ulp
{
GLM_FUNC_QUALIFIER float next_float(float const & x)
{
return GLM_NEXT_AFTER(x, std::numeric_limits<float>::max());
return GLM_NEXT_AFTER_FLT(x, std::numeric_limits<float>::max());
}
GLM_FUNC_QUALIFIER double next_float(double const & x)
{
return GLM_NEXT_AFTER(x, std::numeric_limits<double>::max());
return GLM_NEXT_AFTER_DBL(x, std::numeric_limits<double>::max());
}
template<typename T, template<typename> class vecType>
@@ -40,12 +47,12 @@ namespace ulp
GLM_FUNC_QUALIFIER float prev_float(float const & x)
{
return GLM_NEXT_AFTER(x, std::numeric_limits<float>::min());
return GLM_NEXT_AFTER_FLT(x, std::numeric_limits<float>::min());
}
GLM_FUNC_QUALIFIER double prev_float(double const & x)
{
return GLM_NEXT_AFTER(x, std::numeric_limits<double>::min());
return GLM_NEXT_AFTER_DBL(x, std::numeric_limits<double>::min());
}
template<typename T, template<typename> class vecType>