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

在Java 8或9中只调用Integer。bitCount。

其他回答

当你写出比特模式时,“黑客的喜悦”比特旋转变得更加清晰。

unsigned int bitCount(unsigned int x)
{
  x = ((x >> 1) & 0b01010101010101010101010101010101)
     + (x       & 0b01010101010101010101010101010101);
  x = ((x >> 2) & 0b00110011001100110011001100110011)
     + (x       & 0b00110011001100110011001100110011); 
  x = ((x >> 4) & 0b00001111000011110000111100001111)
     + (x       & 0b00001111000011110000111100001111); 
  x = ((x >> 8) & 0b00000000111111110000000011111111)
     + (x       & 0b00000000111111110000000011111111); 
  x = ((x >> 16)& 0b00000000000000001111111111111111)
     + (x       & 0b00000000000000001111111111111111); 
  return x;
}

第一步将偶数位加到奇数位上,产生每两个位的和。其他步骤将高阶数据块添加到低阶数据块,将数据块的大小一直增加一倍,直到最终计数占用整个int。

几个悬而未决的问题:-

如果这个数是负的呢? 如果这个数字是1024,那么“迭代除以2”方法将迭代10次。

我们可以修改算法以支持负数:-

count = 0
while n != 0
if ((n % 2) == 1 || (n % 2) == -1
    count += 1
  n /= 2  
return count

现在为了克服第二个问题,我们可以编写这样的算法:-

int bit_count(int num)
{
    int count=0;
    while(num)
    {
        num=(num)&(num-1);
        count++;
    }
    return count;
}

完整参考请参见:

http://goursaha.freeoda.com/Miscellaneous/IntegerBitCount.html

这里有一个到目前为止还没有提到的解决方案,使用位字段。下面的程序使用4种不同的方法对100000000个16位整数数组中的设置位进行计数。计时结果在括号中给出(在MacOSX上,使用gcc -O3):

#include <stdio.h>
#include <stdlib.h>

#define LENGTH 100000000

typedef struct {
    unsigned char bit0 : 1;
    unsigned char bit1 : 1;
    unsigned char bit2 : 1;
    unsigned char bit3 : 1;
    unsigned char bit4 : 1;
    unsigned char bit5 : 1;
    unsigned char bit6 : 1;
    unsigned char bit7 : 1;
} bits;

unsigned char sum_bits(const unsigned char x) {
    const bits *b = (const bits*) &x;
    return b->bit0 + b->bit1 + b->bit2 + b->bit3 \
         + b->bit4 + b->bit5 + b->bit6 + b->bit7;
}

int NumberOfSetBits(int i) {
    i = i - ((i >> 1) & 0x55555555);
    i = (i & 0x33333333) + ((i >> 2) & 0x33333333);
    return (((i + (i >> 4)) & 0x0F0F0F0F) * 0x01010101) >> 24;
}

#define out(s) \
    printf("bits set: %lu\nbits counted: %lu\n", 8*LENGTH*sizeof(short)*3/4, s);

int main(int argc, char **argv) {
    unsigned long i, s;
    unsigned short *x = malloc(LENGTH*sizeof(short));
    unsigned char lut[65536], *p;
    unsigned short *ps;
    int *pi;

    /* set 3/4 of the bits */
    for (i=0; i<LENGTH; ++i)
        x[i] = 0xFFF0;

    /* sum_bits (1.772s) */
    for (i=LENGTH*sizeof(short), p=(unsigned char*) x, s=0; i--; s+=sum_bits(*p++));
    out(s);

    /* NumberOfSetBits (0.404s) */
    for (i=LENGTH*sizeof(short)/sizeof(int), pi=(int*)x, s=0; i--; s+=NumberOfSetBits(*pi++));
    out(s);

    /* populate lookup table */
    for (i=0, p=(unsigned char*) &i; i<sizeof(lut); ++i)
        lut[i] = sum_bits(p[0]) + sum_bits(p[1]);

    /* 256-bytes lookup table (0.317s) */
    for (i=LENGTH*sizeof(short), p=(unsigned char*) x, s=0; i--; s+=lut[*p++]);
    out(s);

    /* 65536-bytes lookup table (0.250s) */
    for (i=LENGTH, ps=x, s=0; i--; s+=lut[*ps++]);
    out(s);

    free(x);
    return 0;
}

虽然位域版本非常可读,但计时结果显示它比NumberOfSetBits()慢了4倍以上。基于查找表的实现仍然要快得多,特别是对于一个65 kB的表。

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));
    }
}

下面是功能优等递归解决方案,它是迄今为止最纯粹的一个(并且可以用于任何位长度!):

template<typename T>
int popcnt(T n)
{
  if (n>0)
    return n&1 + popcnt(n>>1);
  return 0; 
}