You are given $n$ numbers $a_1, a_2, \dots, a_n$. You can perform at most $k$ operations. For each operation, you can multiply one of the numbers by $x$. We want to make $a_1\mid a_2\mid \dots\mid a_n$ as large as possible, where $\mid$ denotes the bitwise OR. Find the maximum possible value of $a_1\mid a_2\mid \dots\mid a_n$ after performing at most $k$ operations optimally.
Input
The first line contains three integers $n$, $ k $ and $ x $ ($1 \le n \le 200000$, $1 \le k \le 10$, $2 \le x \le 8$).
The second line contains $n$ integers $a_1, a_2, \dots, a_n$ ($0 \le a_i \le 10^9$).
Output the maximum value of a bitwise OR of sequence elements after performing operations.
3 1 2
1 1 1
3
4 2 3
1 2 4 8
79
For the first sample, any possible choice of doing one operation will result the same three numbers 1, 1, 2 so the result is .
For the second sample if we multiply 8 by 3 two times we'll get 72. In this case the numbers will become 1, 2, 4,72 so the OR value will be 79 and is the largest possible result.
Solution
注意到 $2\le x\le 8$,意味着:每次不论选那个数乘以 $x$,这个数的二进制位数都会增加。
(其实这不算什么Key Observation,当 $x>1$ 时,$x$ 乘任何正整数,该数的二进制位数都会增加)
所以要取得最大值,$k$ 次必然都是乘以同一个数,结果的位数就是这个数最后的位数
所以算法是:
将输入数组从大到小排序,枚举乘 $x^k$ 后长度最长的数,更新答案。
为此,预处理出输入数组前缀和后缀取或(|)的结果。
Implementation
#include <bits/stdc++.h>
using namespace std;
const int N(2e5+);
typedef long long ll;
typedef pair<int,int> P;
P a[N];
int f[N], b[N];
int high_bit(ll x){
for(int i=; i>=; i--){
if(x&(ll)<<i) return i;
}
}
int main(){
int n, k, s;
scanf("%d%d%d", &n, &k, &s);
for(int i=, x; i<=n; i++) scanf("%d", &x), a[i]={x, i}; f[]=;
for(int i=; i<=n; i++) f[i]=f[i-]|a[i].first;
b[n+]=;
for(int i=n; i; i--) b[i]=b[i+]|a[i].first; sort(a+, a+n+, greater<P>());
if(!a[].first){puts(""); return ;}
ll _=; for(int i=; i<k; i++) _*=s; ll ans=, lar=_*a[].first;
int id, h=high_bit(lar);
for(int i=; i<=n; i++){
lar=_*a[i].first;
if(high_bit(lar)<h) break;
id=a[i].second;
ans=max(ans, f[id-]|b[id+]|lar);
}
printf("%lld\n", ans);
}
比赛时把
typedef long long ll;
写成了
typedef long ll;
结果交上去连样例都没过,但在自己电脑上却没问题, 这是由于 long / long long 的具体长度依赖于机器 (machine-dependent),
C++ 标准只规定了其最小长度(minimum size),long(32 bits)long long(64 bits).
除去这个细节不谈,比赛时还犯了个算法上的错误:
枚举原来最长的数而不是乘 $x^k$ 后最长的数。
UPD 2018/3/18
难道不是用最大的数乘 $x^k$ 吗?