#pragma GCC target ("avx2")
#pragma GCC optimization ("O3")
#pragma GCC optimization ("unroll-loops")
#include <bits/stdc++.h>
#define DEBUG 1
using namespace std;
namespace output{
void __(short x){cout<<x;}
void __(unsigned x){cout<<x;}
void __(int x){cout<<x;}
void __(long long x){cout<<x;}
void __(unsigned long long x){cout<<x;}
void __(double x){cout<<x;}
void __(long double x){cout<<x;}
void __(char x){cout<<x;}
void __(const char*x){cout<<x;}
void __(const string&x){cout<<x;}
void __(bool x){cout<<(x?"true":"false");}
template<class S,class T>
void __(const pair<S,T>&x){__(DEBUG?"(":""),__(x.first),__(DEBUG?", ":" "),__(x.second),__(DEBUG?")":"");}
template<class T>
void __(const vector<T>&x){__(DEBUG?"{":"");bool _=0;for(const auto&v:x)__(_?DEBUG?", ":" ":""),__(v),_=1;__(DEBUG?"}":"");}
template<class T>
void __(const set<T>&x){__(DEBUG?"{":"");bool _=0;for(const auto&v:x)__(_?DEBUG?", ":" ":""),__(v),_=1;__(DEBUG?"}":"");}
template<class T>
void __(const multiset<T>&x){__(DEBUG?"{":"");bool _=0;for(const auto&v:x)__(_?DEBUG?", ":" ":""),__(v),_=1;__(DEBUG?"}":"");}
template<class S,class T>
void __(const map<S,T>&x){__(DEBUG?"{":"");bool _=0;for(const auto&v:x)__(_?DEBUG?", ":" ":""),__(v),_=1;__(DEBUG?"}":"");}
void pr(){cout<<"\n";}
template<class S,class... T>
void pr(const S&a,const T&...b){__(a);if(sizeof...(b))__(' ');pr(b...);}
}
using namespace output;
typedef long long ll;
typedef long double ld;
typedef pair<int,int> pii;
typedef pair<ll,ll> pll;
typedef pair<int,char> pic;
typedef pair<double,double> pdd;
typedef pair<ld,ld> pld;
typedef vector<int> vi;
typedef vector<ll> vl;
typedef pair<int,vi> piv;
#define pb push_back
#define fox(i,x,y) for(int i=(x);i<=(y);i++)
#define foxr(i,x,y) for(int i=(y);i>=(x);i--)
const int MN = 3003;
string s, t;
int n, m, i, j, pre[MN], suf[MN], st[MN], lim1, lim2, ans, sw, fl, k, f[MN];
pii sna;
void solve(){
k = lim2;
for(int i=2,j=1;i<=k;i++){
while(j&&t[m-i]!=t[m-j]) j=f[j-1];
f[i]=j++;
}
f[0]=-1;
int ln = 0;
for(int i=n;i>=1;i--){
while(ln>=0&&(ln>k||t[m-ln-1]!=s[i-1])) ln=f[ln];
suf[i]=++ln;
ln=min(ln,k);
st[i]=max(0,k-suf[i]);
}
k = lim1;
for(int i=2,j=1;i<=k;i++){
while(j&&t[i-1]!=t[j-1]) j=f[j-1];
f[i]=j++;
}
ln = 0;
for(int i=1;i<=n;i++){
while(ln>=0&&(ln>k||t[ln]!=s[i-1])) ln=f[ln];
pre[i]=++ln;
ln=min(ln,k);
if(pre[i]+suf[i+1]>ans){
ans = pre[i]+suf[i+1];
if(sw) sna = {n-(i+suf[i+1]), st[i+1]};
else sna = {i-pre[i], st[i+1]};
}
}
}
int main(){
cin >> s >> t;
n = s.size(), m = t.size();
lim1 = 0; lim2 = m;
solve();
for(int i=1;i<=m;i++){
lim1++; lim2--;
char ch = t.back(); t.erase(t.end()-1);
t.insert(t.begin(),ch);
solve();
}
reverse(s.begin(),s.end());
lim1 = 0; lim2 = m; sw = 1;
solve();
for(int i=1;i<=m;i++){
lim1++; lim2--;
char ch = t.back(); t.erase(t.end()-1);
t.insert(t.begin(),ch);
solve();
}
printf("%d\n%d %d\n",ans,sna.first,sna.second);
return 0;
}
Compilation message
necklace.cpp:2:0: warning: ignoring #pragma GCC optimization [-Wunknown-pragmas]
#pragma GCC optimization ("O3")
necklace.cpp:3:0: warning: ignoring #pragma GCC optimization [-Wunknown-pragmas]
#pragma GCC optimization ("unroll-loops")
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
5 ms |
256 KB |
Output is correct |
2 |
Correct |
5 ms |
376 KB |
Output is correct |
3 |
Correct |
5 ms |
376 KB |
Output is correct |
4 |
Correct |
5 ms |
256 KB |
Output is correct |
5 |
Correct |
5 ms |
376 KB |
Output is correct |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
5 ms |
256 KB |
Output is correct |
2 |
Correct |
5 ms |
376 KB |
Output is correct |
3 |
Correct |
5 ms |
376 KB |
Output is correct |
4 |
Correct |
5 ms |
256 KB |
Output is correct |
5 |
Correct |
5 ms |
376 KB |
Output is correct |
6 |
Correct |
9 ms |
376 KB |
Output is correct |
7 |
Correct |
9 ms |
376 KB |
Output is correct |
8 |
Correct |
8 ms |
376 KB |
Output is correct |
9 |
Correct |
8 ms |
376 KB |
Output is correct |
10 |
Correct |
9 ms |
376 KB |
Output is correct |
11 |
Correct |
9 ms |
376 KB |
Output is correct |
12 |
Correct |
9 ms |
380 KB |
Output is correct |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
5 ms |
256 KB |
Output is correct |
2 |
Correct |
5 ms |
376 KB |
Output is correct |
3 |
Correct |
5 ms |
376 KB |
Output is correct |
4 |
Correct |
5 ms |
256 KB |
Output is correct |
5 |
Correct |
5 ms |
376 KB |
Output is correct |
6 |
Correct |
9 ms |
376 KB |
Output is correct |
7 |
Correct |
9 ms |
376 KB |
Output is correct |
8 |
Correct |
8 ms |
376 KB |
Output is correct |
9 |
Correct |
8 ms |
376 KB |
Output is correct |
10 |
Correct |
9 ms |
376 KB |
Output is correct |
11 |
Correct |
9 ms |
376 KB |
Output is correct |
12 |
Correct |
9 ms |
380 KB |
Output is correct |
13 |
Correct |
234 ms |
416 KB |
Output is correct |
14 |
Correct |
188 ms |
376 KB |
Output is correct |
15 |
Correct |
208 ms |
376 KB |
Output is correct |
16 |
Correct |
177 ms |
376 KB |
Output is correct |
17 |
Correct |
237 ms |
376 KB |
Output is correct |
18 |
Correct |
242 ms |
376 KB |
Output is correct |
19 |
Correct |
205 ms |
376 KB |
Output is correct |
20 |
Correct |
207 ms |
376 KB |
Output is correct |
21 |
Correct |
212 ms |
420 KB |
Output is correct |