我正在寻找最快的方法来获得π的值,作为一个个人挑战。更具体地说,我使用的方法不涉及使用#define常量M_PI,或硬编码的数字。

下面的程序测试了我所知道的各种方法。从理论上讲,内联汇编版本是最快的选择,尽管显然不能移植。我将它作为一个基准,与其他版本进行比较。在我的测试中,使用内置函数,4 * atan(1)版本在GCC 4.2上是最快的,因为它自动将atan(1)折叠成一个常量。通过指定-fno-builtin, atan2(0, -1)版本是最快的。

下面是主要的测试程序(pitimes.c):

#include <math.h>
#include <stdio.h>
#include <time.h>

#define ITERS 10000000
#define TESTWITH(x) {                                                       \
    diff = 0.0;                                                             \
    time1 = clock();                                                        \
    for (i = 0; i < ITERS; ++i)                                             \
        diff += (x) - M_PI;                                                 \
    time2 = clock();                                                        \
    printf("%s\t=> %e, time => %f\n", #x, diff, diffclock(time2, time1));   \
}

static inline double
diffclock(clock_t time1, clock_t time0)
{
    return (double) (time1 - time0) / CLOCKS_PER_SEC;
}

int
main()
{
    int i;
    clock_t time1, time2;
    double diff;

    /* Warmup. The atan2 case catches GCC's atan folding (which would
     * optimise the ``4 * atan(1) - M_PI'' to a no-op), if -fno-builtin
     * is not used. */
    TESTWITH(4 * atan(1))
    TESTWITH(4 * atan2(1, 1))

#if defined(__GNUC__) && (defined(__i386__) || defined(__amd64__))
    extern double fldpi();
    TESTWITH(fldpi())
#endif

    /* Actual tests start here. */
    TESTWITH(atan2(0, -1))
    TESTWITH(acos(-1))
    TESTWITH(2 * asin(1))
    TESTWITH(4 * atan2(1, 1))
    TESTWITH(4 * atan(1))

    return 0;
}

内联汇编的东西(fldpi.c)只适用于x86和x64系统:

double
fldpi()
{
    double pi;
    asm("fldpi" : "=t" (pi));
    return pi;
}

和一个构建脚本,构建我正在测试的所有配置(build.sh):

#!/bin/sh
gcc -O3 -Wall -c           -m32 -o fldpi-32.o fldpi.c
gcc -O3 -Wall -c           -m64 -o fldpi-64.o fldpi.c

gcc -O3 -Wall -ffast-math  -m32 -o pitimes1-32 pitimes.c fldpi-32.o
gcc -O3 -Wall              -m32 -o pitimes2-32 pitimes.c fldpi-32.o -lm
gcc -O3 -Wall -fno-builtin -m32 -o pitimes3-32 pitimes.c fldpi-32.o -lm
gcc -O3 -Wall -ffast-math  -m64 -o pitimes1-64 pitimes.c fldpi-64.o -lm
gcc -O3 -Wall              -m64 -o pitimes2-64 pitimes.c fldpi-64.o -lm
gcc -O3 -Wall -fno-builtin -m64 -o pitimes3-64 pitimes.c fldpi-64.o -lm

除了在各种编译器标志之间进行测试(我也比较了32位和64位,因为优化是不同的),我还尝试切换测试的顺序。但是,atan2(0, -1)版本在每次测试中仍然名列前茅。


当前回答

在编译时用D计算PI。

(摘自DSource.org)

/** Calculate pi at compile time
 *
 * Compile with dmd -c pi.d
 */
module calcpi;

import meta.math;
import meta.conv;

/** real evaluateSeries!(real x, real metafunction!(real y, int n) term)
 *
 * Evaluate a power series at compile time.
 *
 * Given a metafunction of the form
 *  real term!(real y, int n),
 * which gives the nth term of a convergent series at the point y
 * (where the first term is n==1), and a real number x,
 * this metafunction calculates the infinite sum at the point x
 * by adding terms until the sum doesn't change any more.
 */
template evaluateSeries(real x, alias term, int n=1, real sumsofar=0.0)
{
  static if (n>1 && sumsofar == sumsofar + term!(x, n+1)) {
     const real evaluateSeries = sumsofar;
  } else {
     const real evaluateSeries = evaluateSeries!(x, term, n+1, sumsofar + term!(x, n));
  }
}

/*** Calculate atan(x) at compile time.
 *
 * Uses the Maclaurin formula
 *  atan(z) = z - z^3/3 + Z^5/5 - Z^7/7 + ...
 */
template atan(real z)
{
    const real atan = evaluateSeries!(z, atanTerm);
}

template atanTerm(real x, int n)
{
    const real atanTerm =  (n & 1 ? 1 : -1) * pow!(x, 2*n-1)/(2*n-1);
}

/// Machin's formula for pi
/// pi/4 = 4 atan(1/5) - atan(1/239).
pragma(msg, "PI = " ~ fcvt!(4.0 * (4*atan!(1/5.0) - atan!(1/239.0))) );

其他回答

在编译时用D计算PI。

(摘自DSource.org)

/** Calculate pi at compile time
 *
 * Compile with dmd -c pi.d
 */
module calcpi;

import meta.math;
import meta.conv;

/** real evaluateSeries!(real x, real metafunction!(real y, int n) term)
 *
 * Evaluate a power series at compile time.
 *
 * Given a metafunction of the form
 *  real term!(real y, int n),
 * which gives the nth term of a convergent series at the point y
 * (where the first term is n==1), and a real number x,
 * this metafunction calculates the infinite sum at the point x
 * by adding terms until the sum doesn't change any more.
 */
template evaluateSeries(real x, alias term, int n=1, real sumsofar=0.0)
{
  static if (n>1 && sumsofar == sumsofar + term!(x, n+1)) {
     const real evaluateSeries = sumsofar;
  } else {
     const real evaluateSeries = evaluateSeries!(x, term, n+1, sumsofar + term!(x, n));
  }
}

/*** Calculate atan(x) at compile time.
 *
 * Uses the Maclaurin formula
 *  atan(z) = z - z^3/3 + Z^5/5 - Z^7/7 + ...
 */
template atan(real z)
{
    const real atan = evaluateSeries!(z, atanTerm);
}

template atanTerm(real x, int n)
{
    const real atanTerm =  (n & 1 ? 1 : -1) * pow!(x, 2*n-1)/(2*n-1);
}

/// Machin's formula for pi
/// pi/4 = 4 atan(1/5) - atan(1/239).
pragma(msg, "PI = " ~ fcvt!(4.0 * (4*atan!(1/5.0) - atan!(1/239.0))) );

为了完整起见,一个c++模板版本,对于一个优化的构建,它将在编译时计算PI的近似值,并将内联到单个值。

#include <iostream>

template<int I>
struct sign
{
    enum {value = (I % 2) == 0 ? 1 : -1};
};

template<int I, int J>
struct pi_calc
{
    inline static double value ()
    {
        return (pi_calc<I-1, J>::value () + pi_calc<I-1, J+1>::value ()) / 2.0;
    }
};

template<int J>
struct pi_calc<0, J>
{
    inline static double value ()
    {
        return (sign<J>::value * 4.0) / (2.0 * J + 1.0) + pi_calc<0, J-1>::value ();
    }
};


template<>
struct pi_calc<0, 0>
{
    inline static double value ()
    {
        return 4.0;
    }
};

template<int I>
struct pi
{
    inline static double value ()
    {
        return pi_calc<I, I>::value ();
    }
};

int main ()
{
    std::cout.precision (12);

    const double pi_value = pi<10>::value ();

    std::cout << "pi ~ " << pi_value << std::endl;

    return 0;
}

注意,对于I > 10,优化构建可能会很慢,对于非优化运行也是如此。对于12次迭代,我相信大约有80k次调用value()(在没有内存的情况下)。

在过去,由于字的大小很小,浮点运算很慢或者根本不存在,我们常常这样做:

/* Return approximation of n * PI; n is integer */
#define pi_times(n) (((n) * 22) / 7)

对于不需要很高精度的应用程序(例如电子游戏),这是非常快速和准确的。

如果你想计算π值的近似值(出于某种原因),你应该尝试二进制提取算法。Bellard对BBP的改进给出了O(N²)中的PI。


如果你想获得π值的近似值来进行计算,那么:

PI = 3.141592654

当然,这只是一个近似值,并不完全准确。误差略大于0.00000000004102。(4个十万亿分之一,大约4/10,000,000,000)。


如果你想用π做数学运算,那就准备好铅笔和纸,或者电脑代数包,然后使用π的精确值π。

如果你真的想要一个公式,这个很有趣:

π = -i ln(-1)

从圆面积计算π:-)

<input id="range" type="range" min="10" max="960" value="10" step="50" oninput="calcPi()"> <br> <div id="cont"></div> <script> function generateCircle(width) { var c = width/2; var delta = 1.0; var str = ""; var xCount = 0; for (var x=0; x <= width; x++) { for (var y = 0; y <= width; y++) { var d = Math.sqrt((x-c)*(x-c) + (y-c)*(y-c)); if (d > (width-1)/2) { str += '.'; } else { xCount++; str += 'o'; } str += "&nbsp;" } str += "\n"; } var pi = (xCount * 4) / (width * width); return [str, pi]; } function calcPi() { var e = document.getElementById("cont"); var width = document.getElementById("range").value; e.innerHTML = "<h4>Generating circle...</h4>"; setTimeout(function() { var circ = generateCircle(width); e.innerHTML = "<pre>" + "π = " + circ[1].toFixed(2) + "\n" + circ[0] +"</pre>"; }, 200); } calcPi(); </script>