Basics

位运算包含如下基本操作: &(and), |(or), ~(not), ^(exclusive-or, xor), << (left shift), >> (right shift)

  • 合并:A | B
  • 交集:A & B
  • 减法:A & ~B
  • 取反:ALL_BITS ^ A or ~A
  • 设置某一位为1:A |= 1 << bit
  • 清除某一位为0:A &= ~(1 << bit)
  • 判断某一位是否为1:(A & 1 << bit) != 0
  • 抽取最后一位bit位:A&-A or A&~(A-1) or x^(x&(x-1))
  • 清除最后一位bit位:A & (A-1)
  • Get all 1-bits ~0 ????

Examples

  • int转换成二进制之后包含的1个数
1
2
3
4
5
6
def countOne(n):
count = 0
while n:
n = n & (n - 1)
count += 1
return count
  • int 求和
1
2
def getSum(a, b):
return a if b == 0 else getSum(a ^ b, (a & b) << 1)
  • int 减法
1
2
def getSubtract(a, b):
return a if b == 0 else getSubtract(a ^ b, (~a & b) << 1)
  • 查找丢失的数字,长度为n的数字,内部包含 0,1,2,…,n,且不重复。现在有一个数字丢失了,找出哪个数字丢失。如n = 4, nums = [0, 1, 3], 丢失的是2
1
2
3
4
5
6
def missingNumber(nums):
ret = 0
for i, num in enumerate(nums):
ret ^= i
ret ^= num
return ret ^ len(nums)
  • 找出小于N的最大 2^x(每次清除最后一位1,找到n变成0之前的最后一个梳子状态也可以)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
def largest_power(n):
n |= n >> 1
n |= n >> 2
n |= n >> 4
n |= n >> 8
n |= n >> 16
return (n + 1) >> 1
def largest_power2(n):
pre = n
while n:
pre = n
n &= n - 1
return pre
  • 给定一个无符号的int,反转对应的二进制
1
2
3
4
5
6
7
def reverseBits2(n):
res = 0
for i in range(32):
res <<= 1
res |= n & 1
n >>= 1
return res
  • 给出区间[m, n],(0 <= m <= n <= 2147483647),返回区间内数字按位and之后的结果;如[5, 7] = 5 & 6 & 7 = 4
1
2
3
4
5
6
7
def rangeBitwiseAnd(m, n):
a = 0
while m != n:
m >>= 1
n >>= 1
a += 1
return m << a
  • 给定无符号的n,转换成二进制之后包含1的数量(Hamming weight)
1
2
3
4
5
6
def hammingWeight(n):
count = 0
while n:
n &= n - 1
count += 1
return count

Application

REPEATED DNA SEQUENCES

在一个DNA字符串(只包含ACGT)中找出重复的长度为10的子串,例:
Given s = “AAAAACCCCCAAAAACCCCCCAAAAAGGGTTT”,
Return: [“AAAAACCCCC”, “CCCCCAAAAA”].

  • 解法-1:遍历所有的子串,通过dict判断是否重复
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
class Solution(object):
def findRepeatedDnaSequences(self, s):
"""
:type s: str
:rtype: List[str]
"""
if len(s) < 10: return []
sub_dict = {}
ans = []
for i in range(len(s) - 10 + 1):
substring = s[i:i + 10]
if sub_dict.has_key(substring):
if sub_dict[substring] == 1:
ans.append(substring)
sub_dict[substring] += 1
else:
sub_dict[substring] = 1
return ans
  • 解法-2:上述需要把所有长度为10的子串都记录下来,会占一定的空间,我们使用二进制来优化空间占用。因为只有ACGT四个字符,所以我们可以考虑用两位二进制来表示一个字符,A = 00,C = 01,G = 10,T = 11,这样对于每个子串,我们可以使用 20位二进制来表示,节省了上述key占有的空间
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
class Solution(object):
def findRepeatedDnaSequences(self, s):
"""
:type s: str
:rtype: List[str]
"""
if len(s) <= 10: return []
char_dict = {'A': 0, 'C': 1, 'G': 2, 'T': 3}
sub_dict = {}
ans = []
cur = 0
for i in range(9):
cur = (cur << 2) | char_dict[s[i]]
for i in range(9, len(s)):
cur = (cur & 0x3ffff) << 2 | char_dict[s[i]]
if sub_dict.has_key(cur):
if sub_dict[cur] == 1:
ans.append(s[i - 9:i + 1])
sub_dict[cur] += 1
else:
sub_dict[cur] = 1
return ans
Majority Element

给定长度为n的数字,找出其中出现次数超过⌊ n/2 ⌋的数字,众数

  • 解法-1:dict记录出现次数,遍历查看出现次数最多的
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
class Solution(object):
def majorityElement(self, nums):
"""
:type nums: List[int]
:rtype: int
"""
num_dict = {}
for i, num in enumerate(nums):
if not num_dict.has_key(num):
num_dict[num] = 0
num_dict[num] += 1
for num in num_dict.keys():
if num_dict[num] >= len(nums) / 2 + len(nums) % 2:
return num
  • 解法-2:迭代32次,每次计算所有n个数第i位为1的个数,由于众数一定存在,所以如果第i位为1的个数大于 ⌊ n/2 ⌋,则众数中相应位肯定也为1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
class Solution(object):
def majorityElement_bit(self, nums):
mask, ret = 1, 0
for i in range(32):
count = 0
for num in nums:
if mask & num: count += 1
if count > len(nums) // 2:
# if the 31th bit if 1,
# it means it's a negative number
if i == 31:
ret = -((1 << 31) - ret)
else:
ret |= mask
mask <<= 1
return ret
Single Number III

给定一组数,其中有两个元素只出现一次,其他都出现两次,找出只出现一次的两个元素

  • 解法-1:dict计数
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
class Solution(object):
def singleNumber(self, nums):
"""
:type nums: List[int]
:rtype: List[int]
"""
num_dict = {}
ans = []
for num in nums:
if not num_dict.has_key(num):
num_dict[num] = 0
num_dict[num] += 1
for num in num_dict.keys():
if num_dict[num] == 1:
ans.append(num)
return ans
  • 解法-2:把所有数字都xor下,可得到a ^ b,通过a ^ b随便取其中等于1的一位(表示a和b在这一位中一个是0,一个是1),去和数组所有的数字 & 下,可以把数组分成两组,且两组数都满足只有一个数字出现一次,其他数字都出现两次。分别对两组数求只出现一次的数字。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
class Solution(object):
def singleNumber_bit(self, nums):
"""
:type nums: List[int]
:rtype: List[int]
"""
s = 0
for x in nums:
s ^= x
s &= -s
ans = [0, 0]
for x in nums:
if x & s:
ans[0] ^= x
else:
ans[1] ^= x
return ans
Maximum Product of Word Lengths

给定一串单词,从中找出两个没有公用字符的单子,且length(word[i]) * length(word[j])最大,返回最大值。

  • 解法:复杂度O(n^2),主要问题是判断两个word是否有公用字符,把使用到的字符转换成二进制数进行记录,两个word对应的二进制数 & 一下如果 = 0,则表示没有公用字符。
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
class Solution(object):
def getBitNum(self, word):
num = 0
for c in word:
num |= 1 << (ord(c) - ord('a'))
return num
def maxProduct(self, words):
"""
:type words: List[str]
:rtype: int
"""
word_bit = {}
ans = 0
for word in words:
word_bit[word] = self.getBitNum(word)
l = len(words)
for i, word1 in enumerate(words):
for j, word2 in enumerate(words, i + 1):
if not word_bit[word1] & word_bit[word2]:
ans = max(ans, len(word1) * len(word2))
return ans

原文:A summary: how to use bit manipulation to solve problems easily and efficiently