haiku/headers/posix/math.h
Axel Dörfler 7f5bbbdc56 * Removed PI, and PI2 from math.h.
* Replaced all occurences with the standard macros M_PI, and M_PI_2.
* Some coding style cleanup on the touched files, no other changes besides
  adding a missing check for a failed memory allocation.


git-svn-id: file:///srv/svn/repos/haiku/haiku/trunk@31250 a95241bf-73f2-0310-859d-f6bbb57e9c96
2009-06-26 11:58:43 +00:00

288 lines
9.0 KiB
C

/*
* Copyright 2001-2009, Haiku.
* Distributed under the terms of the MIT License.
*/
#ifndef _MATH_H_
#define _MATH_H_
#define M_E 2.7182818284590452354 /* e */
#define M_LOG2E 1.4426950408889634074 /* log 2e */
#define M_LOG10E 0.43429448190325182765 /* log 10e */
#define M_LN2 0.69314718055994530942 /* log e2 */
#define M_LN10 2.30258509299404568402 /* log e10 */
#define M_PI 3.14159265358979323846 /* pi */
#define M_PI_2 1.57079632679489661923 /* pi/2 */
#define M_PI_4 0.78539816339744830962 /* pi/4 */
#define M_1_PI 0.31830988618379067154 /* 1/pi */
#define M_2_PI 0.63661977236758134308 /* 2/pi */
#define M_2_SQRTPI 1.12837916709551257390 /* 2/sqrt(pi) */
#define M_SQRT2 1.41421356237309504880 /* sqrt(2) */
#define M_SQRT1_2 0.70710678118654752440 /* 1/sqrt(2) */
/* platform independent IEEE floating point special values */
#define __HUGE_VAL_v 0x7ff0000000000000LL
#define __huge_val_t union { unsigned char __c[8]; long long __ll; double __d; }
#ifndef HUGE_VAL
# define HUGE_VAL (((__huge_val_t) { __ll: __HUGE_VAL_v }).__d)
#endif
#define __HUGE_VALF_v 0x7f800000L
#define __huge_valf_t union { unsigned char __c[4]; long __l; float __f; }
#define HUGE_VALF (((__huge_valf_t) { __l: __HUGE_VALF_v }).__f)
/* TODO: define HUGE_VALL for long doubles */
#define __NAN_VALF_v 0x7fc00000L
#define NAN (((__huge_valf_t) { __l: __NAN_VALF_v }).__f)
#define INFINITY HUGE_VALF
/* floating-point categories */
#define FP_NAN 0
#define FP_INFINITE 1
#define FP_ZERO 2
#define FP_SUBNORMAL 3
#define FP_NORMAL 4
#ifdef __cplusplus
struct __exception;
extern "C" int matherr(struct __exception *);
struct __exception {
#else
struct exception;
extern int matherr(struct exception *);
struct exception {
#endif
int type;
char *name;
double arg1;
double arg2;
double retval;
};
#define DOMAIN 1
#define SING 2
#define OVERFLOW 3
#define UNDERFLOW 4
#define TLOSS 5
#define PLOSS 6
extern int signgam;
#ifdef __cplusplus
extern "C" {
#endif
/* float math functions */
extern float acosf(float x);
extern float acoshf(float x);
extern float asinf(float x);
extern float asinhf(float x);
extern float atan2f(float y, float x);
extern float atanf(float x);
extern float atanhf(float x);
extern float cbrtf(float x);
extern float ceilf(float x);
extern float cosf(float x);
extern float coshf(float x);
extern float expf(float x);
extern float expm1f(float x);
extern float fabsf(float x);
extern float floorf(float x);
extern float fmodf(float x, float y);
extern float frexpf(float x, int *_exponent);
extern float gammaf(float x);
extern float hypotf(float x, float y);
extern float ldexpf(float x, int exponent);
extern float lgammaf(float x);
extern long long llrintf(float x);
extern float log10f(float x);
extern float log1pf(float x);
extern float logbf(float x);
extern float logf(float x);
extern long lrintf(float x);
extern long lroundf(float x);
extern float modff(float x, float *y);
extern float nearbyintf(float x);
extern float powf(float x, float y);
extern float remquof(float x, float y, int *quo);
extern float roundf(float x);
extern float sinf(float x);
extern float sinhf(float x);
extern float sqrtf(float x);
extern float tanf(float x);
extern float tanhf(float x);
/* double math functions */
extern double acos(double x);
extern double acosh(double x);
extern double asin(double x);
extern double asinh(double x);
extern double atan(double x);
extern double atan2(double x, double y);
extern double atanh(double x);
extern double ceil(double x);
extern double cos(double x);
extern double cosh(double x);
extern double exp(double x);
extern double fabs(double x);
extern double floor(double x);
extern double fmod(double x, double y);
extern double frexp(double x, int *_exponent);
extern double gamma(double x);
extern double ldexp(double x, int exponent);
extern double lgamma(double x);
extern long long llrint(double x);
extern double log(double x);
extern double log10(double x);
extern long lrint(double x);
extern long lround(double x);
extern double modf(double x, double *y);
extern double nearbyint(double x);
extern double pow(double x, double y);
extern double remquo(double x, double y, int *quo);
extern double round(double x);
extern double sin(double x);
extern double sinh(double x);
extern double sqrt(double x);
extern double tan(double x);
extern double tanh(double x);
extern double trunc(double x);
/* long double math functions */
extern long double acosl(long double x);
extern long double acoshl(long double x);
extern long double asinl(long double x);
extern long double atanl(long double x);
extern long double atanhl(long double x);
extern long double atan2l(long double y, long double x);
extern long double lgammal(long double x);
extern long double nearbyintl(long double x);
extern long long llrintl(long double x);
extern long lrintl(long double x);
extern long lroundl(long double x);
extern long double remquol(long double x, long double y, int *quo);
extern long double roundl(long double x);
/* some BSD non-ANSI or POSIX math functions */
extern double cbrt(double x);
extern double erf(double x);
extern double erfc(double x);
extern double expm1(double x);
extern double gamma_r(double x, int *y);
extern double hypot(double x, double y);
extern int ilogb(double x);
extern double j0(double x);
extern double j1(double x);
extern double jn(int x, double y);
extern double lgamma_r(double x, int *y);
extern double log1p(double x);
extern double logb(double x);
extern double nextafter(double x, double y);
extern double nexttoward(double x, long double y);
extern double remainder(double x, double y);
extern double rint(double x);
extern double scalb (double x, double y);
extern double y0(double x);
extern double y1(double x);
extern double yn(int x, double y);
/* other stuff as defined in BeOS */
extern int isinff(float value);
extern int finitef(float value);
extern float dremf(float x, float y);
extern float significandf(float x);
extern float copysignf(float x, float y);
extern int isnanf(float value);
extern double significand(double x);
extern double copysign(double x, double y);
extern double scalbln(double x, long n);
extern double scalbn(double x, int y);
extern double drem(double x, double y);
extern int finite(double x);
extern float j0f(float x);
extern float j1f(float x);
extern float jnf(int x, float y);
extern float y0f(float x);
extern float y1f(float x);
extern float ynf(int x, float y);
extern float erff(float x);
extern float erfcf(float x);
extern float gammaf_r(float x, int *y);
extern float lgammaf_r(float x, int *y);
extern float rintf(float x);
extern float nextafterf(float x, float y);
extern float nexttowardf(float x, long double y);
extern float remainderf(float x, float y);
extern float scalbf(float x, float n);
extern float scalbnf(float x, int n);
extern int ilogbf(float x);
extern long double nextafterl(long double x, long double y);
extern long double nexttowardl(long double x, long double y);
extern long double remainderl(long double x, long double y);
extern long double rintl(long double x);
extern long double scalbnl(long double x, int n);
extern long double scalblnl(long double x, long n);
/* prototypes for functions used in the macros below */
extern int __fpclassifyf(float value);
extern int __signbitf(float value);
extern int __finitef(float value);
extern int __isnanf(float value);
extern int __isinff(float value);
extern int __fpclassifyl(long double value);
extern int __signbitl(long double value);
extern int __finitel(long double value);
extern int __isnanl(long double value);
extern int __isinfl(long double value);
extern int __fpclassify(double value);
extern int __signbit(double value);
extern int __finite(double value);
extern int __isnan(double value);
extern int __isinf(double value);
/* returns number of classification appropriate for 'value' */
#define fpclassify(value) \
(sizeof(value) == sizeof(float) ? __fpclassifyf(value) \
: sizeof(value) == sizeof(double) ? __fpclassify(value) \
: __fpclassifyl(value))
/* returns non-zero if 'value' is negative */
# define signbit(value) \
(sizeof(value) == sizeof(float) ? __signbitf(value) \
: sizeof(value) == sizeof(double) ? __signbit(value) \
: __signbitl(value))
/* returns non-zero if 'value' is not Inf or NaN */
# define isfinite(value) \
(sizeof(value) == sizeof(float) ? __finitef(value) \
: sizeof(value) == sizeof(double) ? __finite(value) \
: __finitel(value))
/* returns non-zero if 'value' is neither zero, sub-normal, Inf, nor NaN */
# define isnormal(value) \
(fpclassify(value) == FP_NORMAL)
/* returns non-zero if 'value' is NaN */
# define isnan(value) \
(sizeof(value) == sizeof(float) ? __isnanf(value) \
: sizeof(value) == sizeof(double) ? __isnan(value) \
: __isnanl(value))
/* returns non-zero if 'value is Inf */
# define isinf(value) \
(sizeof(value) == sizeof(float) ? __isinff(value) \
: sizeof(value) == sizeof(double) ? __isinf(value) \
: __isinfl(value))
#ifdef __cplusplus
}
#endif
#endif /* _MATH_H_ */