이 제출은 이전 버전의 oj.uz에서 채점하였습니다. 현재는 제출 당시와는 다른 서버에서 채점을 하기 때문에, 다시 제출하면 결과가 달라질 수도 있습니다.
#include "factories.h"
#include <bits/stdc++.h>
#define taskname ""
#define all(x) x.begin(), x.end()
#define rall(x) x.rbegin(), x.rend()
#define ll long long
#define ld long double
#define pb push_back
#define ff first
#define ss second
#define pii pair<int, int>
#define vi vector<int>
#define vll vector<ll>
#define vvi vector<vi>
#define vii vector<pii>
#define isz(x) (int)x.size()
using namespace std;
const int mxN = 1e6 + 10;
const ll oo = 1e18;
int n;
vector<vii> G;
vi vtn;
vector<vector<pair<ll, ll>>> vt;
ll dis[mxN];
pii sparse[20][mxN];
int timer, ecnt, ap[mxN], dep[mxN], cur[mxN], tin[mxN], tout[mxN];
void dfs(int v, int p)
{
tin[v] = ++timer;
if (p == -1){
dep[v] = dis[v] = 0;
}else dep[v] = dep[p] + 1;
sparse[0][++ecnt] = {dep[v], v};
ap[v] = ecnt;
for (auto &[u, w] : G[v])
{
if (u == p)
continue;
dis[u] = dis[v] + w;
dfs(u, v);
sparse[0][++ecnt] = {dep[v], v};
}
tout[v] = ++timer;
}
void build_sparse()
{
for (int p = 1, i = 1; p << 1 <= ecnt; p <<= 1, ++i)
for (int j = 1; j <= ecnt - (p << 1) + 1; ++j)
sparse[i][j] = min(sparse[i - 1][j], sparse[i - 1][j + p]);
}
bool is_ancestor(int u, int v)
{
return tin[u] <= tin[v] && tout[u] >= tout[v];
}
int LCA(int u, int v)
{
int l = ap[u], r = ap[v];
if (l > r)
swap(l, r);
int len = r - l + 1, lg_len = __lg(len);
return min(sparse[lg_len][l], sparse[lg_len][r - (1 << lg_len) + 1]).ss;
}
void Init(int N, int A[], int B[], int D[])
{
n = N;
G.resize(n);
vt.resize(n);
for (int i = 0; i < N - 1; ++i)
{
G[A[i]].emplace_back(B[i], D[i]);
G[B[i]].emplace_back(A[i], D[i]);
}
dfs(0, -1);
build_sparse();
}
long long Query(int S, int X[], int T, int Y[])
{
// queue<pii> q;
for (int i = 0; i < S; ++i)
{
cur[X[i]] = 1;
// q.emplace(X[i], 0);
vtn.emplace_back(X[i]);
}
for (int i = 0; i < T; ++i)
{
cur[Y[i]] = 2;
vtn.emplace_back(Y[i]);
}
sort(all(vtn), [&](int u, int v)
{ return tin[u] < tin[v]; });
for (int i = 0; i < S + T - 1; ++i)
{
vtn.emplace_back(LCA(vtn[i], vtn[i + 1]));
}
sort(all(vtn), [&](int u, int v)
{ return tin[u] < tin[v]; });
vtn.erase(unique(all(vtn)), vtn.end());
auto add_vt = [&](int u, int v, ll w)
{
vt[u].emplace_back(v, w);
vt[v].emplace_back(u, w);
};
stack<int> s;
for (int i = 0; i < isz(vtn); ++i)
{
while (not s.empty() && not is_ancestor(s.top(), vtn[i]))
s.pop();
if (s.empty())
s.emplace(vtn[i]);
else
{
add_vt(s.top(), vtn[i], dis[vtn[i]] - dis[s.top()]);
s.emplace(vtn[i]);
}
}
ll res = oo;
auto dfs = [&](auto self, int v, int p) -> vll
{
vll cur_dis(2, oo);
if (cur[v])
cur_dis[cur[v] - 1] = 0;
for (auto &[u, w] : vt[v])
{
if (u == p)
continue;
vll tmp = self(self, u, v);
for (int i = 0; i < 2; ++i)
cur_dis[i] = min(cur_dis[i], tmp[i] + w);
}
res = min(res, accumulate(all(cur_dis), 0LL));
return cur_dis;
};
dfs(dfs, vtn[0], -1);
// for (int i = 0; i < S; ++i)
// {
// cur[X[i]] = 0;
// // vt[X[i]].clear();
// }
// for (int i = 0; i < T; ++i)
// {
// cur[Y[i]] = 0;
// // vt[Y[i]].clear();
// }
for (int i:vtn) vt[i].clear(), cur[i] = 0;
vtn.clear();
return res;
}
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |
# | Verdict | Execution time | Memory | Grader output |
---|
Fetching results... |