ruby--ruby/math.c

530 lines
9.4 KiB
C

/**********************************************************************
math.c -
$Author$
$Date$
created at: Tue Jan 25 14:12:56 JST 1994
Copyright (C) 1993-2003 Yukihiro Matsumoto
**********************************************************************/
#include "ruby.h"
#include <math.h>
#include <errno.h>
VALUE rb_mMath;
#define Need_Float(x) (x) = rb_Float(x)
#define Need_Float2(x,y) do {\
Need_Float(x);\
Need_Float(y);\
} while (0)
static void
domain_check(x, msg)
double x;
char *msg;
{
while(1) {
if (errno) {
rb_sys_fail(msg);
}
if (isnan(x)) {
#if defined(EDOM)
errno = EDOM;
#elif defined(ERANGE)
errno = ERANGE;
#endif
continue;
}
break;
}
}
/*
* call-seq:
* Math.atan2(y, x) => float
*
* Computes the arc tangent given <i>y</i> and <i>x</i>. Returns
* -PI..PI.
*
*/
static VALUE
math_atan2(obj, y, x)
VALUE obj, x, y;
{
Need_Float2(y, x);
return rb_float_new(atan2(RFLOAT(y)->value, RFLOAT(x)->value));
}
/*
* call-seq:
* Math.cos(x) => float
*
* Computes the cosine of <i>x</i> (expressed in radians). Returns
* -1..1.
*/
static VALUE
math_cos(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(cos(RFLOAT(x)->value));
}
/*
* call-seq:
* Math.sin(x) => float
*
* Computes the sine of <i>x</i> (expressed in radians). Returns
* -1..1.
*/
static VALUE
math_sin(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(sin(RFLOAT(x)->value));
}
/*
* call-seq:
* Math.tan(x) => float
*
* Returns the tangent of <i>x</i> (expressed in radians).
*/
static VALUE
math_tan(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(tan(RFLOAT(x)->value));
}
/*
* call-seq:
* Math.acos(x) => float
*
* Computes the arc cosine of <i>x</i>. Returns 0..PI.
*/
static VALUE
math_acos(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = acos(RFLOAT(x)->value);
domain_check(d, "acos");
return rb_float_new(d);
}
/*
* call-seq:
* Math.asin(x) => float
*
* Computes the arc sine of <i>x</i>. Returns 0..PI.
*/
static VALUE
math_asin(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = asin(RFLOAT(x)->value);
domain_check(d, "asin");
return rb_float_new(d);
}
/*
* call-seq:
* Math.atan(x) => float
*
* Computes the arc tangent of <i>x</i>. Returns -{PI/2} .. {PI/2}.
*/
static VALUE
math_atan(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(atan(RFLOAT(x)->value));
}
#ifndef HAVE_COSH
double
cosh(x)
double x;
{
return (exp(x) + exp(-x)) / 2;
}
#endif
/*
* call-seq:
* Math.cosh(x) => float
*
* Computes the hyperbolic cosine of <i>x</i> (expressed in radians).
*/
static VALUE
math_cosh(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(cosh(RFLOAT(x)->value));
}
#ifndef HAVE_SINH
double
sinh(x)
double x;
{
return (exp(x) - exp(-x)) / 2;
}
#endif
/*
* call-seq:
* Math.sinh(x) => float
*
* Computes the hyperbolic sine of <i>x</i> (expressed in
* radians).
*/
static VALUE
math_sinh(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(sinh(RFLOAT(x)->value));
}
#ifndef HAVE_TANH
double
tanh(x)
double x;
{
return sinh(x) / cosh(x);
}
#endif
/*
* call-seq:
* Math.tanh() => float
*
* Computes the hyperbolic tangent of <i>x</i> (expressed in
* radians).
*/
static VALUE
math_tanh(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(tanh(RFLOAT(x)->value));
}
/*
* call-seq:
* Math.acosh(x) => float
*
* Computes the inverse hyperbolic cosine of <i>x</i>.
*/
static VALUE
math_acosh(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = acosh(RFLOAT(x)->value);
domain_check(d, "acosh");
return rb_float_new(d);
}
/*
* call-seq:
* Math.asinh(x) => float
*
* Computes the inverse hyperbolic sine of <i>x</i>.
*/
static VALUE
math_asinh(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(asinh(RFLOAT(x)->value));
}
/*
* call-seq:
* Math.atanh(x) => float
*
* Computes the inverse hyperbolic tangent of <i>x</i>.
*/
static VALUE
math_atanh(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = atanh(RFLOAT(x)->value);
domain_check(d, "atanh");
return rb_float_new(d);
}
/*
* call-seq:
* Math.exp(x) => float
*
* Returns e**x.
*/
static VALUE
math_exp(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(exp(RFLOAT(x)->value));
}
#if defined __CYGWIN__
# include <cygwin/version.h>
# if CYGWIN_VERSION_DLL_MAJOR < 1005
# define nan(x) nan()
# endif
# define log(x) ((x) < 0.0 ? nan("") : log(x))
# define log10(x) ((x) < 0.0 ? nan("") : log10(x))
#endif
/*
* call-seq:
* Math.log(numeric) => float
*
* Returns the natural logarithm of <i>numeric</i>.
*/
static VALUE
math_log(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = log(RFLOAT(x)->value);
domain_check(d, "log");
return rb_float_new(d);
}
/*
* call-seq:
* Math.log10(numeric) => float
*
* Returns the base 10 logarithm of <i>numeric</i>.
*/
static VALUE
math_log10(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = log10(RFLOAT(x)->value);
domain_check(d, "log10");
return rb_float_new(d);
}
/*
* call-seq:
* Math.sqrt(numeric) => float
*
* Returns the non-negative square root of <i>numeric</i>.
*/
static VALUE
math_sqrt(obj, x)
VALUE obj, x;
{
double d;
Need_Float(x);
errno = 0;
d = sqrt(RFLOAT(x)->value);
domain_check(d, "sqrt");
return rb_float_new(d);
}
/*
* call-seq:
* Math.frexp(numeric) => [ fraction, exponent ]
*
* Returns a two-element array containing the normalized fraction (a
* <code>Float</code>) and exponent (a <code>Fixnum</code>) of
* <i>numeric</i>.
*
* fraction, exponent = Math.frexp(1234) #=> [0.6025390625, 11]
* fraction * 2**exponent #=> 1234.0
*/
static VALUE
math_frexp(obj, x)
VALUE obj, x;
{
double d;
int exp;
Need_Float(x);
d = frexp(RFLOAT(x)->value, &exp);
return rb_assoc_new(rb_float_new(d), INT2NUM(exp));
}
/*
* call-seq:
* Math.ldexp(flt, int) -> float
*
* Returns the value of <i>flt</i>*(2**<i>int</i>).
*
* fraction, exponent = Math.frexp(1234)
* Math.ldexp(fraction, exponent) #=> 1234.0
*/
static VALUE
math_ldexp(obj, x, n)
VALUE obj, x, n;
{
Need_Float(x);
return rb_float_new(ldexp(RFLOAT(x)->value, NUM2INT(n)));
}
/*
* call-seq:
* Math.hypot(x, y) => float
*
* Returns sqrt(x**2 + y**2), the hypotenuse of a right-angled triangle
* with sides <i>x</i> and <i>y</i>.
*
* Math.hypot(3, 4) #=> 5.0
*/
static VALUE
math_hypot(obj, x, y)
VALUE obj, x, y;
{
Need_Float2(x, y);
return rb_float_new(hypot(RFLOAT(x)->value, RFLOAT(y)->value));
}
/*
* call-seq:
* Math.erf(x) => float
*
* Calculates the error function of x.
*/
static VALUE
math_erf(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(erf(RFLOAT(x)->value));
}
/*
* call-seq:
* Math.erfc(x) => float
*
* Calculates the complementary error function of x.
*/
static VALUE
math_erfc(obj, x)
VALUE obj, x;
{
Need_Float(x);
return rb_float_new(erfc(RFLOAT(x)->value));
}
/*
* The <code>Math</code> module contains module functions for basic
* trigonometric and transcendental functions. See class
* <code>Float</code> for a list of constants that
* define Ruby's floating point accuracy.
*/
void
Init_Math()
{
rb_mMath = rb_define_module("Math");
#ifdef M_PI
rb_define_const(rb_mMath, "PI", rb_float_new(M_PI));
#else
rb_define_const(rb_mMath, "PI", rb_float_new(atan(1.0)*4.0));
#endif
#ifdef M_E
rb_define_const(rb_mMath, "E", rb_float_new(M_E));
#else
rb_define_const(rb_mMath, "E", rb_float_new(exp(1.0)));
#endif
rb_define_module_function(rb_mMath, "atan2", math_atan2, 2);
rb_define_module_function(rb_mMath, "cos", math_cos, 1);
rb_define_module_function(rb_mMath, "sin", math_sin, 1);
rb_define_module_function(rb_mMath, "tan", math_tan, 1);
rb_define_module_function(rb_mMath, "acos", math_acos, 1);
rb_define_module_function(rb_mMath, "asin", math_asin, 1);
rb_define_module_function(rb_mMath, "atan", math_atan, 1);
rb_define_module_function(rb_mMath, "cosh", math_cosh, 1);
rb_define_module_function(rb_mMath, "sinh", math_sinh, 1);
rb_define_module_function(rb_mMath, "tanh", math_tanh, 1);
rb_define_module_function(rb_mMath, "acosh", math_acosh, 1);
rb_define_module_function(rb_mMath, "asinh", math_asinh, 1);
rb_define_module_function(rb_mMath, "atanh", math_atanh, 1);
rb_define_module_function(rb_mMath, "exp", math_exp, 1);
rb_define_module_function(rb_mMath, "log", math_log, 1);
rb_define_module_function(rb_mMath, "log10", math_log10, 1);
rb_define_module_function(rb_mMath, "sqrt", math_sqrt, 1);
rb_define_module_function(rb_mMath, "frexp", math_frexp, 1);
rb_define_module_function(rb_mMath, "ldexp", math_ldexp, 2);
rb_define_module_function(rb_mMath, "hypot", math_hypot, 2);
rb_define_module_function(rb_mMath, "erf", math_erf, 1);
rb_define_module_function(rb_mMath, "erfc", math_erfc, 1);
}