Defined in header <cmath> | ||
---|---|---|
Rounding to floating-point types | ||
(1) | ||
float rint ( float num ); double rint ( double num ); long double rint ( long double num ); | (since C++11) (until C++23) | |
/* floating-point-type */ rint( /* floating-point-type */ num ); | (since C++23) | |
float rintf( float num ); | (2) | (since C++11) |
long double rintl( long double num ); | (3) | (since C++11) |
long | ||
(4) | ||
long lrint ( float num ); long lrint ( double num ); long lrint ( long double num ); | (since C++11) (until C++23) | |
long lrint ( /* floating-point-type */ num ); | (since C++23) | |
long lrintf( float num ); | (5) | (since C++11) |
long lrintl( long double num ); | (6) | (since C++11) |
long long | ||
(7) | ||
long long llrint ( float num ); long long llrint ( double num ); long long llrint ( long double num ); | (since C++11) (until C++23) | |
long long llrint ( /* floating-point-type */ num ); | (since C++23) | |
long long llrintf( float num ); | (8) | (since C++11) |
long long llrintl( long double num ); | (9) | (since C++11) |
Additional overloads | ||
Defined in header <cmath> | ||
template< class Integer > double rint( Integer num ); | (A) | (since C++11) |
template< class Integer > long lrint( Integer num ); | (B) | (since C++11) |
template< class Integer > long long llrint( Integer num ); | (C) | (since C++11) |
num
to an integer value (in floating-point format), using the current rounding mode. The library provides overloads of std::rint
for all cv-unqualified floating-point types as the type of the parameter num
. (since C++23)
num
to an integer value, using the current rounding mode. The library provides overloads of std::lrint
and std::llrint
for all cv-unqualified floating-point types as the type of the parameter num
. (since C++23)
double
num | - | floating-point or integer value |
If no errors occur, the nearest integer value to num
, according to the current rounding mode, is returned.
Errors are reported as specified in math_errhandling
.
If the result of std::lrint
or std::llrint
is outside the range representable by the return type, a domain error or a range error may occur.
If the implementation supports IEEE floating-point arithmetic (IEC 60559), For the std::rint
function:
num
is ±∞, it is returned, unmodified num
is ±0, it is returned, unmodified num
is NaN, NaN is returned std::lrint
and std::llrint
functions: num
is ±∞, FE_INVALID
is raised and an implementation-defined value is returned FE_INVALID
is raised and an implementation-defined value is returned num
is NaN, FE_INVALID
is raised and an implementation-defined value is returned POSIX specifies that all cases where std::lrint
or std::llrint
raise FE_INEXACT
are domain errors.
As specified in math_errhandling
, FE_INEXACT
may be (but isn't required to be on non-IEEE floating-point platforms) raised by std::rint
when rounding a non-integer finite value.
The only difference between std::rint
and std::nearbyint
is that std::nearbyint
never raises FE_INEXACT
.
The largest representable floating-point values are exact integers in all standard floating-point formats, so std::rint
never overflows on its own; however the result may overflow any integer type (including std::intmax_t
), when stored in an integer variable.
If the current rounding mode is:
FE_DOWNWARD
, then std::rint
is equivalent to std::floor
. FE_UPWARD
, then std::rint
is equivalent to std::ceil
. FE_TOWARDZERO
, then std::rint
is equivalent to std::trunc
FE_TONEAREST
, then std::rint
differs from std::round
in that halfway cases are rounded to even rather than away from zero. The additional overloads are not required to be provided exactly as (A-C). They only need to be sufficient to ensure that for their argument num
of integer type:
std::rint(num)
has the same effect as std::rint(static_cast<double>(num))
. std::lrint(num)
has the same effect as std::lrint(static_cast<double>(num))
. std::llrint(num)
has the same effect as std::llrint(static_cast<double>(num))
. #include <cfenv> #include <climits> #include <cmath> #include <iostream> // #pragma STDC FENV_ACCESS ON int main() { std::fesetround(FE_TONEAREST); std::cout << "Rounding to nearest (halfway cases to even):\n" << " rint(+2.3) = " << std::rint(2.3) << '\n' << " rint(+2.5) = " << std::rint(2.5) << '\n' << " rint(+3.5) = " << std::rint(3.5) << '\n' << " rint(-2.3) = " << std::rint(-2.3) << '\n' << " rint(-2.5) = " << std::rint(-2.5) << '\n' << " rint(-3.5) = " << std::rint(-3.5) << '\n'; std::fesetround(FE_DOWNWARD); std::cout << "Rounding down:\n" << " rint(+2.3) = " << std::rint(2.3) << '\n' << " rint(+2.5) = " << std::rint(2.5) << '\n' << " rint(+3.5) = " << std::rint(3.5) << '\n' << " rint(-2.3) = " << std::rint(-2.3) << '\n' << " rint(-2.5) = " << std::rint(-2.5) << '\n' << " rint(-3.5) = " << std::rint(-3.5) << '\n' << "Rounding down with lrint:\n" << " lrint(+2.3) = " << std::lrint(2.3) << '\n' << " lrint(+2.5) = " << std::lrint(2.5) << '\n' << " lrint(+3.5) = " << std::lrint(3.5) << '\n' << " lrint(-2.3) = " << std::lrint(-2.3) << '\n' << " lrint(-2.5) = " << std::lrint(-2.5) << '\n' << " lrint(-3.5) = " << std::lrint(-3.5) << '\n' << "Special values:\n" << " lrint(-0.0) = " << std::lrint(-0.0) << '\n' << std::hex << std::showbase << " lrint(-Inf) = " << std::lrint(-INFINITY) << '\n'; // error handling std::feclearexcept(FE_ALL_EXCEPT); std::cout << "std::rint(0.1) = " << std::rint(.1) << '\n'; if (std::fetestexcept(FE_INEXACT)) std::cout << " FE_INEXACT was raised\n"; std::feclearexcept(FE_ALL_EXCEPT); std::cout << "std::lrint(LONG_MIN-2048.0) = " << std::lrint(LONG_MIN - 2048.0) << '\n'; if (std::fetestexcept(FE_INVALID)) std::cout << " FE_INVALID was raised\n"; }
Possible output:
Rounding to nearest (halfway cases to even): rint(+2.3) = 2 rint(+2.5) = 2 rint(+3.5) = 4 rint(-2.3) = -2 rint(-2.5) = -2 rint(-3.5) = -4 Rounding down: rint(+2.3) = 2 rint(+2.5) = 2 rint(+3.5) = 4 rint(-2.3) = -2 rint(-2.5) = -2 rint(-3.5) = -4 Rounding down with lrint: lrint(+2.3) = 2 lrint(+2.5) = 2 lrint(+3.5) = 3 lrint(-2.3) = -3 lrint(-2.5) = -3 lrint(-3.5) = -4 Special values: lrint(-0.0) = 0 lrint(-Inf) = 0x8000000000000000 std::rint(0.1) = 0 std::lrint(LONG_MIN-2048.0) = 0x8000000000000000 FE_INVALID was raised
(C++11)(C++11)(C++11) | nearest integer not greater in magnitude than the given value (function) |
(C++11)(C++11)(C++11) | nearest integer using current rounding mode (function) |
(C++11)(C++11) | gets or sets rounding direction (function) |
C documentation for rint |
© cppreference.com
Licensed under the Creative Commons Attribution-ShareAlike Unported License v3.0.
https://en.cppreference.com/w/cpp/numeric/math/rint