Each of these macro constants expands to a nonnegative integer constant expression, which can be used with std::fesetround and std::fegetround to indicate one of the supported floating-point rounding modes. The implementation may define additional rounding mode constants in <cfenv>, which should all begin with FE_ followed by at least one uppercase letter. Each macro is only defined if it is supported.

Additional rounding modes may be supported by an implementation.

The current rounding mode affects the following:

• results of floating-point arithmetic operators outside of constant expressions

double x = 1;
x / 10; // 0.09999999999999999167332731531132594682276248931884765625
     // or 0.1000000000000000055511151231257827021181583404541015625

• results of standard library mathematical functions

std::sqrt(2); // 1.41421356237309492343001693370752036571502685546875
           // or 1.4142135623730951454746218587388284504413604736328125

• floating-point to floating-point implicit conversion and casts

double d = 1 + std::numeric_limits<double>::epsilon();
float f = d; // 1.00000000000000000000000
          // or 1.00000011920928955078125

• string conversions such as std::strtod or std::printf

std::stof("0.1"); // 0.0999999940395355224609375
               // or 0.100000001490116119384765625

• the library rounding functions std::nearbyint, std::rint, std::lrint

std::lrint(2.1); // 2 or 3

The current rounding mode does NOT affect the following:

• floating-point to integer implicit conversion and casts (always towards zero),
• results of floating-point arithmetic operators in expressions executed at compile time (always to nearest),
• the library functions std::round, std::lround, std::llround, std::ceil, std::floor, std::trunc.

As with any floating-point environment functionality, rounding is only guaranteed if #pragma STDC FENV_ACCESS ON is set.

Compilers that do not support the pragma may offer their own ways to support current rounding mode. For example Clang and GCC have the option -frounding-math intended to disable optimizations that would change the meaning of rounding-sensitive code.

Example

#include <cfenv>
#include <cmath>
#include <iomanip>
#include <iostream>
#include <string>
// #pragma STDC FENV_ACCESS ON

int main()
{
    std::fesetround(FE_DOWNWARD);
    std::cout << "rounding down: \n" << std::setprecision(50)
              << "         pi = " << std::acos(-1.f) << '\n'
              << "stof(\"1.1\") = " << std::stof("1.1") << '\n'
              << "  rint(2.1) = " << std::rint(2.1) << "\n\n";
    std::fesetround(FE_UPWARD);
    std::cout << "rounding up: \n"
              << "         pi = " << std::acos(-1.f) << '\n'
              << "stof(\"1.1\") = " << std::stof("1.1") << '\n'
              << "  rint(2.1) = " << std::rint(2.1) << '\n';
}

rounding down:
         pi = 3.141592502593994140625
stof("1.1") = 1.099999904632568359375
  rint(2.1) = 2

rounding up:
         pi = 3.1415927410125732421875
stof("1.1") = 1.10000002384185791015625
  rint(2.1) = 3

See also