2011-04-26 11:29:56 +00:00
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// OpenGL Mathematics Copyright (c) 2005 - 2011 G-Truc Creation (www.g-truc.net)
|
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
// Created : 2011-04-26
|
|
|
|
// Updated : 2011-04-26
|
|
|
|
// Licence : This source is under MIT licence
|
|
|
|
// File : test/gtx/ulp.cpp
|
|
|
|
///////////////////////////////////////////////////////////////////////////////////////////////////
|
|
|
|
|
|
|
|
#include <glm/glm.hpp>
|
|
|
|
#include <glm/gtx/ulp.hpp>
|
|
|
|
#include <iostream>
|
|
|
|
|
|
|
|
int test_ulp_float()
|
|
|
|
{
|
2011-04-26 13:46:30 +00:00
|
|
|
double Temp = nextafter(0.02, 0.01);
|
|
|
|
std::cout << Temp << " 0.01, 0.02" << std::endl;
|
|
|
|
|
2011-04-26 13:24:45 +00:00
|
|
|
std::size_t A = glm::ulp(0.01, 0.02);
|
|
|
|
std::size_t B = glm::ulp(glm::vec2(0.01), glm::vec2(0.02));
|
|
|
|
std::size_t C = glm::ulp(glm::vec3(0.01), glm::vec3(0.02));
|
|
|
|
std::size_t D = glm::ulp(glm::vec4(0.01), glm::vec4(0.02));
|
2011-04-26 11:29:56 +00:00
|
|
|
std::cout << "glm::ulp test: " << A << std::endl;
|
|
|
|
std::cout << "glm::ulp test: " << B << std::endl;
|
|
|
|
std::cout << "glm::ulp test: " << C << std::endl;
|
|
|
|
std::cout << "glm::ulp test: " << D << std::endl;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int main()
|
|
|
|
{
|
2011-04-26 13:46:30 +00:00
|
|
|
std::cout << "Test 76" << std::endl;
|
2011-04-26 11:29:56 +00:00
|
|
|
test_ulp_float();
|
|
|
|
}
|
|
|
|
|
|
|
|
|