mirror of
https://github.com/g-truc/glm.git
synced 2024-11-30 03:44:38 +00:00
97 lines
1.8 KiB
C++
97 lines
1.8 KiB
C++
#include <glm/ext/scalar_ulp.hpp>
|
|
#include <glm/ext/scalar_relational.hpp>
|
|
|
|
static int test_ulp_float_dist()
|
|
{
|
|
int Error = 0;
|
|
|
|
float A = 1.0f;
|
|
|
|
float B = glm::nextFloat(A);
|
|
Error += glm::notEqual(A, B, 0) ? 0 : 1;
|
|
float C = glm::prevFloat(B);
|
|
Error += glm::equal(A, C, 0) ? 0 : 1;
|
|
|
|
int D = glm::floatDistance(A, B);
|
|
Error += D == 1 ? 0 : 1;
|
|
int E = glm::floatDistance(A, C);
|
|
Error += E == 0 ? 0 : 1;
|
|
|
|
return Error;
|
|
}
|
|
|
|
static int test_ulp_float_step()
|
|
{
|
|
int Error = 0;
|
|
|
|
float A = 1.0f;
|
|
|
|
for(int i = 10; i < 1000; i *= 10)
|
|
{
|
|
float B = glm::nextFloat(A, i);
|
|
Error += glm::notEqual(A, B, 0) ? 0 : 1;
|
|
float C = glm::prevFloat(B, i);
|
|
Error += glm::equal(A, C, 0) ? 0 : 1;
|
|
|
|
int D = glm::floatDistance(A, B);
|
|
Error += D == i ? 0 : 1;
|
|
int E = glm::floatDistance(A, C);
|
|
Error += E == 0 ? 0 : 1;
|
|
}
|
|
|
|
return Error;
|
|
}
|
|
|
|
static int test_ulp_double_dist()
|
|
{
|
|
int Error = 0;
|
|
|
|
double A = 1.0;
|
|
|
|
double B = glm::nextFloat(A);
|
|
Error += glm::notEqual(A, B, 0) ? 0 : 1;
|
|
double C = glm::prevFloat(B);
|
|
Error += glm::equal(A, C, 0) ? 0 : 1;
|
|
|
|
glm::int64 const D = glm::floatDistance(A, B);
|
|
Error += D == 1 ? 0 : 1;
|
|
glm::int64 const E = glm::floatDistance(A, C);
|
|
Error += E == 0 ? 0 : 1;
|
|
|
|
return Error;
|
|
}
|
|
|
|
static int test_ulp_double_step()
|
|
{
|
|
int Error = 0;
|
|
|
|
double A = 1.0;
|
|
|
|
for(int i = 10; i < 1000; i *= 10)
|
|
{
|
|
double B = glm::nextFloat(A, i);
|
|
Error += glm::notEqual(A, B, 0) ? 0 : 1;
|
|
double C = glm::prevFloat(B, i);
|
|
Error += glm::equal(A, C, 0) ? 0 : 1;
|
|
|
|
glm::int64 const D = glm::floatDistance(A, B);
|
|
Error += D == i ? 0 : 1;
|
|
glm::int64 const E = glm::floatDistance(A, C);
|
|
Error += E == 0 ? 0 : 1;
|
|
}
|
|
|
|
return Error;
|
|
}
|
|
|
|
int main()
|
|
{
|
|
int Error = 0;
|
|
|
|
Error += test_ulp_float_dist();
|
|
Error += test_ulp_float_step();
|
|
Error += test_ulp_double_dist();
|
|
Error += test_ulp_double_step();
|
|
|
|
return Error;
|
|
}
|