//#pragma GCC target ("avx2")
#pragma GCC optimization ("O3")
#pragma GCC optimization ("unroll-loops")
#include <bits/stdc++.h>
#define owo(i,a, b) for(auto i=(a);i<(b); ++i)
#define uwu(i,a, b) for(auto i=(a)-1; i>=(b); --i)
#define senpai push_back
#define ttgl pair<int, int>
#define ayaya cout<<"ayaya~"<<endl
using namespace std;
/*#include <ext/pb_ds/assoc_container.hpp>
#include <ext/pb_ds/tree_policy.hpp>
using namespace __gnu_pbds;
#define ordered_set tree<ttgl, null_type,less<ttgl>, rb_tree_tag,tree_order_statistics_node_update>*/
using ll = long long;
using ld = long double;
const ll MOD = 1000000007;
const ll root = 62;
ll gcd(ll a,ll b){return b?gcd(b,a%b):a;}
ll binpow(ll a,ll b){ll res=1;while(b){if(b&1)res=(res*a)%MOD;a=(a*a)%MOD;b>>=1;}return res;}
ll modInv(ll a){return binpow(a, MOD-2);}
const double PI = acos(-1);
const double eps = 1e-10;
const int INF = 0x3f3f3f3f;
const int NINF = 0xc0c0c0c0;
const ll INFLL = 0x3f3f3f3f3f3f3f3f;
const ll NINFLL = 0xc0c0c0c0c0c0c0c0;
const int mxN = 100001;
int n, m;
//maintain the cursor positon and multiple fenwick trees
struct BIT{
int bit[2*mxN];
void add(int p, int x) {
p += mxN;
while(p < 2*mxN) {
bit[p] += x;
p += p&-p;
}
}
void upd(int l, int r, int x) {
if(l > r) swap(l, r);
add(l, x);
add(r + 1, -x);
}
int sum(int p) {
int res = 0;
p += mxN;
while(p) {
res += bit[p];
p -= p&-p;
}
return res;
}
};
BIT f1[2], f2[2];
int curr[2];
int arr[mxN][2];
int pos[2];
int del[2];
int main() {
//freopen("file.in", "r", stdin);
//freopen("file.out", "w", stdout);
mt19937_64 rng(chrono::steady_clock::now().time_since_epoch().count());
cin.tie(0)->sync_with_stdio(0);
cin>>n;
pos[0] = 1;
pos[1] = 1;
owo(i, 0, n) {
cin>>arr[i][0]>>arr[i][1];
if(i) {
owo(j, 0, 2) {
f2[j].upd(pos[j], pos[j] + arr[i][j], 1);
pos[j] += arr[i][j];
}
}else {
owo(j, 0, 2) {
f1[j].upd(pos[j], pos[j] + arr[i][j], 1);
pos[j] += arr[i][j];
}
}
}
cin>>m;
char z;
int a, b;
pos[0] = 1;
pos[1] = 1;
int T = 0;
owo(i, 0, m) {
cin>>z;
if(z=='Q') {
cout<<(f1[0].sum(0) + f1[1].sum(0) + f2[0].sum(del[0]) + f2[1].sum(del[1]))<<"\n";
}
if(z=='B') {
if(T) {
owo(j, 0, 2) {
f1[j].upd(pos[j], pos[j] + arr[T][j], -1);
f2[j].upd(pos[j] + del[j], pos[j] + arr[T][j] + del[j], 1);
pos[j] -= arr[T-1][j];
}
T--;
}
}
if(z=='F') {
if(T < n-1) {
T++;
owo(j, 0, 2) {
pos[j] += arr[T-1][j];
f1[j].upd(pos[j], pos[j] + arr[T][j], 1);
f2[j].upd(pos[j] + del[j], pos[j] + arr[T][j] + del[j], -1);
}
}
}
if(z=='C') {
cin>>a>>b;
del[0] += arr[T][0] - a;
f1[0].upd(pos[0], pos[0] + arr[T][0], -1);
f1[0].upd(pos[0], pos[0] + a, 1);
del[1] += arr[T][1] - b;
f1[1].upd(pos[1], pos[1] + arr[T][1], -1);
f1[1].upd(pos[1], pos[1] + b, 1);
arr[T][0] = a;
arr[T][1] = b;
}
}
return 0;
}
Compilation message
ruka.cpp:2: warning: ignoring #pragma GCC optimization [-Wunknown-pragmas]
2 | #pragma GCC optimization ("O3")
|
ruka.cpp:3: warning: ignoring #pragma GCC optimization [-Wunknown-pragmas]
3 | #pragma GCC optimization ("unroll-loops")
|
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
1 ms |
512 KB |
Output is correct |
2 |
Correct |
1 ms |
512 KB |
Output is correct |
3 |
Correct |
1 ms |
512 KB |
Output is correct |
4 |
Correct |
1 ms |
512 KB |
Output is correct |
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
1 ms |
512 KB |
Output is correct |
2 |
Correct |
1 ms |
512 KB |
Output is correct |
3 |
Correct |
1 ms |
512 KB |
Output is correct |
4 |
Correct |
1 ms |
512 KB |
Output is correct |
5 |
Correct |
41 ms |
2612 KB |
Output is correct |
6 |
Correct |
35 ms |
2168 KB |
Output is correct |
7 |
Correct |
43 ms |
2296 KB |
Output is correct |
8 |
Correct |
43 ms |
2424 KB |
Output is correct |
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
1 ms |
512 KB |
Output is correct |
2 |
Correct |
1 ms |
512 KB |
Output is correct |
3 |
Correct |
1 ms |
512 KB |
Output is correct |
4 |
Correct |
1 ms |
512 KB |
Output is correct |
5 |
Correct |
41 ms |
2612 KB |
Output is correct |
6 |
Correct |
35 ms |
2168 KB |
Output is correct |
7 |
Correct |
43 ms |
2296 KB |
Output is correct |
8 |
Correct |
43 ms |
2424 KB |
Output is correct |
9 |
Correct |
98 ms |
4600 KB |
Output is correct |
10 |
Execution timed out |
2052 ms |
5112 KB |
Time limit exceeded |
11 |
Halted |
0 ms |
0 KB |
- |