cbrt, cbrtf, cbrtl
From cppreference.com
C
Concurrency support (C11)
Common mathematical functions
Functions
Basic operations
Maximum/minimum operations
(C23)
(C23)
(C23)
(C23)
(C23)
Exponential functions
Power functions
Trigonometric and hyperbolic functions
Nearest integer floating-point
Floating-point manipulation
Narrowing operations
Quantum and quantum exponent
(C23)
(C23)
(C23)
(C23)
Decimal re-encoding functions
(C23)
(C23)
(C23)
(C23)
Total order and payload functions
(C23)
(C23)
(C23)
(C23)
Classification
(C99)
(C23)
(C99)
(C99)
(C99)
(C99)
(C99)
(C23)
(C23)
(C99)
(C99)
(C99)
(C99)
(C99)
(C99)
(C23)
(C23)
Error and gamma functions
Types
(C99)(C99)
(C99)(C99)
(C23)(C23)
Macro constants
Special floating-point values
(C99)(C99)(C23)
(C99)(C23)
(C99)(C23)
Arguments and return values
(C99)(C99)
(C99)(C99)(C99)(C99)(C99)
(C23)(C23)
(C23)(C23)(C23)(C23)(C23)
Error handling
(C99)(C99)
(C99)
Fast operation indicators
(C99)(C99)
(C23)(C23)(C23)(C23)
(C23)(C23)(C23)(C23)
(C23)(C23)(C23)(C23)
(C99)(C23)
(C23)(C23)(C23)(C23)
(C23)(C23)(C23)(C23)
(C23)(C23)(C23)(C23)
Defined in header
<math.h>
float cbrtf( float arg );
(1)
(since C99)
double cbrt( double arg );
(2)
(since C99)
long double cbrtl( long double arg );
(3)
(since C99)
Defined in header
<tgmath.h>
#define cbrt( arg )
(4)
(since C99)
1-3) Computes the cube root of
arg
.4) Type-generic macro: If
arg
has type long double, cbrtl
is called. Otherwise, if arg
has integer type or the type double, cbrt
is called. Otherwise, cbrtf
is called.[edit] Parameters
arg
-
floating-point value
[edit] Return value
If no errors occur, the cube root of arg
(\(\small{\sqrt[3]{arg} }\)3√arg), is returned.
If a range error occurs due to underflow, the correct result (after rounding) is returned.
[edit] Error handling
Errors are reported as specified in math_errhandling.
If the implementation supports IEEE floating-point arithmetic (IEC 60559),
- if the argument is ±0 or ±∞, it is returned, unchanged
- if the argument is NaN, NaN is returned.
[edit] Notes
cbrt(arg) is not equivalent to pow (arg, 1.0/3) because the rational number \(\small{\frac1{3} }\) 1
3
is typically not equal to 1.0/3 and std::pow cannot raise a negative base to a fractional exponent. Moreover, cbrt(arg) usually gives more accurate results than pow (arg, 1.0/3) (see example).
[edit] Example
Run this code
#include <float.h> #include <math.h> #include <stdio.h> int main(void) { printf ("Normal use:\n" "cbrt(729) = %f\n", cbrt(729)); printf ("cbrt(-0.125) = %f\n", cbrt(-0.125)); printf ("Special values:\n" "cbrt(-0) = %f\n", cbrt(-0.0)); printf ("cbrt(+inf) = %f\n", cbrt(INFINITY)); printf ("Accuracy:\n" "cbrt(343) = %.*f\n", DBL_DECIMAL_DIG, cbrt(343)); printf ("pow(343,1.0/3) = %.*f\n", DBL_DECIMAL_DIG, pow (343, 1.0/3)); }
Possible output:
Normal use: cbrt(729) = 9.000000 cbrt(-0.125) = -0.500000 Special values: cbrt(-0) = -0.000000 cbrt(+inf) = inf Accuracy: cbrt(343) = 7.00000000000000000 pow(343,1.0/3) = 6.99999999999999911
[edit] References
- C23 standard (ISO/IEC 9899:2024):
- 7.12.7.1 The cbrt functions (p: TBD)
- 7.25 Type-generic math <tgmath.h> (p: TBD)
- F.10.4.1 The cbrt functions (p: TBD)
- C17 standard (ISO/IEC 9899:2018):
- 7.12.7.1 The cbrt functions (p: 180-181)
- 7.25 Type-generic math <tgmath.h> (p: 272-273)
- F.10.4.1 The cbrt functions (p: 381-)
- C11 standard (ISO/IEC 9899:2011):
- 7.12.7.1 The cbrt functions (p: 247)
- 7.25 Type-generic math <tgmath.h> (p: 373-375)
- F.10.4.1 The cbrt functions (p: 524)
- C99 standard (ISO/IEC 9899:1999):
- 7.12.7.1 The cbrt functions (p: 228)
- 7.22 Type-generic math <tgmath.h> (p: 335-337)
- F.9.4.1 The cbrt functions (p: 460)
[edit] See also
(C99)(C99)(C99)
+y2
)
(function) [edit]
C++ documentation for cbrt