代表数字7的8位像这样:

00000111

设置了三个比特。

确定32位整数中设置位数的算法是什么?


当前回答

如果你使用c++,另一个选择是使用模板元编程:

// recursive template to sum bits in an int
template <int BITS>
int countBits(int val) {
        // return the least significant bit plus the result of calling ourselves with
        // .. the shifted value
        return (val & 0x1) + countBits<BITS-1>(val >> 1);
}

// template specialisation to terminate the recursion when there's only one bit left
template<>
int countBits<1>(int val) {
        return val & 0x1;
}

用法如下:

// to count bits in a byte/char (this returns 8)
countBits<8>( 255 )

// another byte (this returns 7)
countBits<8>( 254 )

// counting bits in a word/short (this returns 1)
countBits<16>( 256 )

当然,你可以进一步扩展这个模板来使用不同的类型(甚至是自动检测位大小),但为了清晰起见,我让它保持简单。

edit:忘了说这很好,因为它应该在任何c++编译器中工作,它基本上只是为你展开循环,如果一个常量值用于比特计数(换句话说,我很确定这是你能找到的最快的通用方法)

其他回答

一个快速的c#解决方案,使用预先计算的字节位计数表,并根据输入大小进行分支。

public static class BitCount
{
    public static uint GetSetBitsCount(uint n)
    {
        var counts = BYTE_BIT_COUNTS;
        return n <= 0xff ? counts[n]
             : n <= 0xffff ? counts[n & 0xff] + counts[n >> 8]
             : n <= 0xffffff ? counts[n & 0xff] + counts[(n >> 8) & 0xff] + counts[(n >> 16) & 0xff]
             : counts[n & 0xff] + counts[(n >> 8) & 0xff] + counts[(n >> 16) & 0xff] + counts[(n >> 24) & 0xff];
    }

    public static readonly uint[] BYTE_BIT_COUNTS =
    {
        0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4,
        1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
        1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
        2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
        1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
        2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
        2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
        3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
        1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5,
        2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
        2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
        3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
        2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6,
        3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
        3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7,
        4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8
    };
}

你可以使用内置函数__builtin_popcount()。c++中没有__builtin_popcount,但它是GCC编译器的内置函数。这个函数返回一个整数中的设置位数。

int __builtin_popcount (unsigned int x);

参考:Bit Twiddling Hacks

int countBits(int x)
{
    int n = 0;
    if (x) do n++;
           while(x=x&(x-1));
    return n;
}   

或者:

int countBits(int x) { return (x)? 1+countBits(x&(x-1)): 0; }

在我最初的回答7年半之后,@PeterMortensen质疑这是否是有效的C语法。我发布了一个在线编译器的链接,显示它实际上是完全有效的语法(代码如下)。

#include <stdio.h>
int countBits(int x)
{
    int n = 0;
    if (x) do n++;           /* Totally Normal Valid code. */
           while(x=x&(x-1)); /* Nothing to see here.       */
    return n;
}   
 
int main(void) {
    printf("%d\n", countBits(25));
    return 0;
}
 

输出:

3

如果你想重新写清楚,它看起来是这样的:

if (x)
{
    do
    {
        n++;
    } while(x=x&(x-1));
}

但在我看来,这太过分了。

然而,我也意识到函数可以变得更短,但可能更神秘,写为:

int countBits(int x)
{
    int n = 0;
    while (x) x=(n++,x&(x-1));
    return n;
}   

对于那些想要在c++ 11中为任何无符号整数类型作为consexpr函数的人(tacklelib/include/tacklelib/utility/math.hpp):

#include <stdint.h>
#include <limits>
#include <type_traits>

const constexpr uint32_t uint32_max = (std::numeric_limits<uint32_t>::max)();

namespace detail
{
    template <typename T>
    inline constexpr T _count_bits_0(const T & v)
    {
        return v - ((v >> 1) & 0x55555555);
    }

    template <typename T>
    inline constexpr T _count_bits_1(const T & v)
    {
        return (v & 0x33333333) + ((v >> 2) & 0x33333333);
    }

    template <typename T>
    inline constexpr T _count_bits_2(const T & v)
    {
        return (v + (v >> 4)) & 0x0F0F0F0F;
    }

    template <typename T>
    inline constexpr T _count_bits_3(const T & v)
    {
        return v + (v >> 8);
    }

    template <typename T>
    inline constexpr T _count_bits_4(const T & v)
    {
        return v + (v >> 16);
    }

    template <typename T>
    inline constexpr T _count_bits_5(const T & v)
    {
        return v & 0x0000003F;
    }

    template <typename T, bool greater_than_uint32>
    struct _impl
    {
        static inline constexpr T _count_bits_with_shift(const T & v)
        {
            return
                detail::_count_bits_5(
                    detail::_count_bits_4(
                        detail::_count_bits_3(
                            detail::_count_bits_2(
                                detail::_count_bits_1(
                                    detail::_count_bits_0(v)))))) + count_bits(v >> 32);
        }
    };

    template <typename T>
    struct _impl<T, false>
    {
        static inline constexpr T _count_bits_with_shift(const T & v)
        {
            return 0;
        }
    };
}

template <typename T>
inline constexpr T count_bits(const T & v)
{
    static_assert(std::is_integral<T>::value, "type T must be an integer");
    static_assert(!std::is_signed<T>::value, "type T must be not signed");

    return uint32_max >= v ?
        detail::_count_bits_5(
            detail::_count_bits_4(
                detail::_count_bits_3(
                    detail::_count_bits_2(
                        detail::_count_bits_1(
                            detail::_count_bits_0(v)))))) :
        detail::_impl<T, sizeof(uint32_t) < sizeof(v)>::_count_bits_with_shift(v);
}

谷歌测试库中的附加测试:

#include <stdlib.h>
#include <time.h>

namespace {
    template <typename T>
    inline uint32_t _test_count_bits(const T & v)
    {
        uint32_t count = 0;
        T n = v;
        while (n > 0) {
            if (n % 2) {
                count += 1;
            }
            n /= 2;
        }
        return count;
    }
}

TEST(FunctionsTest, random_count_bits_uint32_100K)
{
    srand(uint_t(time(NULL)));
    for (uint32_t i = 0; i < 100000; i++) {
        const uint32_t r = uint32_t(rand()) + (uint32_t(rand()) << 16);
        ASSERT_EQ(_test_count_bits(r), count_bits(r));
    }
}

TEST(FunctionsTest, random_count_bits_uint64_100K)
{
    srand(uint_t(time(NULL)));
    for (uint32_t i = 0; i < 100000; i++) {
        const uint64_t r = uint64_t(rand()) + (uint64_t(rand()) << 16) + (uint64_t(rand()) << 32) + (uint64_t(rand()) << 48);
        ASSERT_EQ(_test_count_bits(r), count_bits(r));
    }
}

我使用下面的函数。我还没有检查基准测试,但它是有效的。

int msb(int num)
{
    int m = 0;
    for (int i = 16; i > 0; i = i>>1)
    {
        // debug(i, num, m);
        if(num>>i)
        {
            m += i;
            num>>=i;
        }
    }
    return m;
}