1c68979fb4
The documentation comment belongs to interface/wx/math.h, not the header under include. Also change the argument type to wxUint32 as the non-gcc version only works for 32 bit values.
111 lines
2.6 KiB
C
111 lines
2.6 KiB
C
/////////////////////////////////////////////////////////////////////////////
|
|
// Name: math.h
|
|
// Purpose: interface of global functions
|
|
// Author: wxWidgets team
|
|
// Licence: wxWindows licence
|
|
/////////////////////////////////////////////////////////////////////////////
|
|
|
|
/** @addtogroup group_funcmacro_math */
|
|
//@{
|
|
|
|
/**
|
|
Returns a non-zero value if @a x is neither infinite nor NaN (not a
|
|
number), returns 0 otherwise.
|
|
|
|
@header{wx/math.h}
|
|
*/
|
|
int wxFinite(double x);
|
|
|
|
/**
|
|
Returns the greatest common divisor of the two given numbers.
|
|
|
|
@since 3.1.0
|
|
|
|
@header{wx/math.h}
|
|
*/
|
|
unsigned int wxGCD(unsigned int u, unsigned int v);
|
|
|
|
/**
|
|
Returns a non-zero value if x is NaN (not a number), returns 0 otherwise.
|
|
|
|
@header{wx/math.h}
|
|
*/
|
|
bool wxIsNaN(double x);
|
|
|
|
/**
|
|
Converts the given array of 10 bytes (corresponding to 80 bits) to
|
|
a float number according to the IEEE floating point standard format
|
|
(aka IEEE standard 754).
|
|
|
|
@see wxConvertToIeeeExtended() to perform the opposite operation
|
|
*/
|
|
wxFloat64 wxConvertFromIeeeExtended(const wxInt8 *bytes);
|
|
|
|
/**
|
|
Converts the given floating number @a num in a sequence of 10 bytes
|
|
which are stored in the given array @a bytes (which must be large enough)
|
|
according to the IEEE floating point standard format
|
|
(aka IEEE standard 754).
|
|
|
|
@see wxConvertFromIeeeExtended() to perform the opposite operation
|
|
*/
|
|
void wxConvertToIeeeExtended(wxFloat64 num, wxInt8 *bytes);
|
|
|
|
/**
|
|
Convert degrees to radians.
|
|
|
|
This function simply returns its argument multiplied by @c M_PI/180 but is
|
|
more readable than writing this expression directly.
|
|
|
|
@see wxRadToDeg()
|
|
|
|
@since 3.1.0
|
|
*/
|
|
double wxDegToRad(double deg);
|
|
|
|
/**
|
|
Convert radians to degrees.
|
|
|
|
This function simply returns its argument multiplied by @c 180/M_PI but is
|
|
more readable than writing this expression directly.
|
|
|
|
@see wxDegToRad()
|
|
|
|
@since 3.1.0
|
|
*/
|
|
double wxRadToDeg(double rad);
|
|
|
|
/**
|
|
Count the number of trailing zeros.
|
|
|
|
This function returns the number of trailing zeros in the binary notation
|
|
of its argument @a x. E.g. for @a x equal to 4, or 0b100, the return value
|
|
is 2.
|
|
|
|
@param x Strictly positive, i.e. non-zero, 32 bit number.
|
|
|
|
@since 3.1.2
|
|
*/
|
|
unsigned int wxCTZ(wxUint32 x);
|
|
|
|
/**
|
|
Small wrapper around round().
|
|
*/
|
|
int wxRound(double x);
|
|
|
|
/**
|
|
Returns true if both double values are identical. This is
|
|
only reliable if both values have been assigned the same
|
|
value.
|
|
*/
|
|
bool wxIsSameDouble(double x, double y);
|
|
|
|
/**
|
|
Return true of @a x is exactly zero. This is only reliable
|
|
if it has been assigned 0.
|
|
*/
|
|
bool wxIsNullDouble(double x);
|
|
|
|
//@}
|
|
|