// Copyright © 2022 Youngmin Park. All rights reserved.
//#pragma GCC optimize("O3")
//#pragma GCC target("avx2")
#include <bits/stdc++.h>
using namespace std;
using ll = long long;
using vi = vector<int>;
using pii = pair<int, int>;
using vpi = vector<pii>;
using pll = pair<ll, ll>;
using vl = vector<ll>;
using vpl = vector<pll>;
using ld = long double;
template <typename T, size_t SZ>
using ar = array<T, SZ>;
template <typename T>
using pqg = priority_queue<T, vector<T>, greater<T>>;
#define all(v) (v).begin(), (v).end()
#define pb push_back
#define sz(x) (int)(x).size()
#define fi first
#define se second
#define lb lower_bound
#define ub upper_bound
constexpr int INF = 1e9;
constexpr ll LINF = 1e18;
const ld PI = acos((ld)-1.0);
constexpr int dx[4] = {1, 0, -1, 0}, dy[4] = {0, 1, 0, -1};
mt19937_64 rng(chrono::steady_clock::now().time_since_epoch().count());
template <typename T>
constexpr bool ckmin(T &a, const T &b) { return b < a ? a = b, 1 : 0; }
template <typename T>
constexpr bool ckmax(T &a, const T &b) { return b > a ? a = b, 1 : 0; }
template <typename A, typename B>
ostream &operator<<(ostream &os, const pair<A, B> &p)
{
return os << '(' << p.first << ", " << p.second << ')';
}
template <typename T_container, typename T = typename enable_if<!is_same<T_container, string>::value, typename T_container::value_type>::type>
ostream &operator<<(ostream &os, const T_container &v)
{
os << '{';
string sep;
for (const T &x : v)
os << sep << x, sep = ", ";
return os << '}';
}
template <typename T>
ostream &operator<<(ostream &os, const deque<T> &v) {
os << vector<T>(all(v));
return os;
}
template <typename T, typename S, typename C>
ostream &operator<<(ostream &os, priority_queue<T, S, C> pq) {
vector<T> v;
while (sz(pq)) {
v.pb(pq.top());
pq.pop();
}
os << v;
return os;
}
void dbg_out()
{
cerr << "\033[0m" << endl;
}
template <typename Head, typename... Tail>
void dbg_out(Head H, Tail... T)
{
cerr << ' ' << H;
dbg_out(T...);
}
#ifdef LOCAL
#define dbg(...) cerr << "\033[1;35m" << __func__ << ':' << __LINE__ << " (" << #__VA_ARGS__ << "):\033[33m", dbg_out(__VA_ARGS__)
#else
#define dbg(...) 42
#endif
inline namespace RecursiveLambda
{
template <typename Fun>
struct y_combinator_result
{
Fun fun_;
template <typename T>
explicit y_combinator_result(T &&fun) : fun_(forward<T>(fun)) {}
template <typename... Args>
decltype(auto) operator()(Args &&...args)
{
return fun_(ref(*this), forward<Args>(args)...);
}
};
template <typename Fun>
decltype(auto) y_combinator(Fun &&fun)
{
return y_combinator_result<decay_t<Fun>>(forward<Fun>(fun));
}
};
inline namespace Range {
class ForwardRange {
int src, dst;
public:
explicit constexpr ForwardRange(const int l, const int r) : src(l), dst(r) {}
explicit constexpr ForwardRange(const int n) : src(0), dst(n) {}
constexpr ForwardRange begin() const { return *this; }
constexpr monostate end() const { return {}; }
constexpr bool operator!=(monostate) const { return src < dst; }
constexpr void operator++() const {}
constexpr int operator*() { return src++; }
};
class BackwardRange {
int src, dst;
public:
explicit constexpr BackwardRange(const int l, const int r) : src(r), dst(l) {}
explicit constexpr BackwardRange(const int n) : src(n), dst(0) {}
constexpr BackwardRange begin() const { return *this; }
constexpr monostate end() const { return {}; }
constexpr bool operator!=(monostate) const { return src > dst; }
constexpr void operator++() const {}
constexpr int operator*() { return --src; }
};
using rep = ForwardRange;
using per = BackwardRange;
};
/**
* Description: point update and rectangle sum with offline 2D BIT.
* For each of the points to be updated, $x\in (0,SZ)$ and $y\neq 0$.
* Time: O(N\log^2 N)
* Memory: O(N\log N)
* Source: Benq
* Verification:
* https://dmoj.ca/problem/occ19g4
* http://www.usaco.org/index.php?page=viewproblem2&cpid=722 (753 ms)
* http://www.usaco.org/index.php?page=viewproblem2&cpid=601 (679 ms)
*/
template<class T, int SZ> struct BITOff2D {
bool mode = 0; // mode = 1 -> initialized
vpi todo; // locations of updates to process
int cnt[SZ], st[SZ];
vi val; vector<T> bit; // store all BITs in single vector
void init() { assert(!mode); mode = 1;
int lst[SZ]; for (int i : rep(SZ)) lst[i] = cnt[i] = 0;
sort(all(todo),[](const pii& a, const pii& b) {
return a.se < b.se; });
for (auto &t : todo) for (int x = t.fi; x < SZ; x += x & -x)
if (lst[x] != t.se) lst[x] = t.se, cnt[x]++;
int sum = 0; for (int i : rep(SZ)) lst[i] = 0, st[i] = (sum += cnt[i]);
val.resize(sum); bit.resize(sum); reverse(all(todo));
for (auto &t : todo) for (int x = t.fi; x < SZ; x += x & -x)
if (lst[x] != t.se) lst[x] = t.se, val[--st[x]] = t.se;
}
int rank(int y, int l, int r) {
return ub(begin(val) + l, begin(val) + r, y) - begin(val) - l; }
void UPD(int x, int y, T t) {
for (y = rank(y, st[x], st[x] + cnt[x]); y <= cnt[x]; y += y & -y)
bit[st[x] + y - 1] += t; }
void upd(int x, int y, T t) {
if (!mode) todo.pb({x, y});
else for (; x < SZ; x += x & -x) UPD(x, y, t); }
int QUERY(int x, int y) { T res = 0;
for (y = rank(y, st[x], st[x] + cnt[x]); y; y -= y & -y) res += bit[st[x] + y - 1];
return res; }
T query(int x, int y) { assert(mode);
T res = 0; for (; x; x -= x & -x) res += QUERY(x, y);
return res; }
T query(int xl, int xr, int yl, int yr) {
return query(xr, yr) - query(xl - 1, yr)
- query(xr, yl - 1) + query(xl - 1, yl - 1); }
};
BITOff2D<int, 210000> bit;
void solve()
{
int n, q;
cin >> n >> q;
vpi sorted_by_sum(n);
vector<ar<int, 3>> queries(q);
vi ind(q), ans(q), xs(n);
for (int i : rep(n)) {
int s, t;
cin >> s >> t;
t++;
sorted_by_sum[i] = {s, t};
xs[i] = s;
// bit.upd(s, t, 1);
}
sort(all(sorted_by_sum), [&](auto x, auto y){
return x.fi + x.se > y.fi + y.se;
});
for (int i : rep(q)) {
auto &[x, y, z] = queries[i];
cin >> x >> y >> z;
y++;
xs.pb(x);
}
iota(all(ind), 0);
sort(all(ind), [&](int i, int j){
return queries[i][2] > queries[j][2];
});
sort(all(xs)), xs.resize(unique(all(xs)) - xs.begin());
for (auto &[s, t] : sorted_by_sum) {
s = lb(all(xs), s) - xs.begin() + 1;
bit.upd(s, t, 1);
}
bit.init();
int ptr{};
for (int i : rep(q)) {
auto& [x, y, z] = queries[ind[i]];
x = lb(all(xs), x) - xs.begin() + 1;
while (ptr < n && xs[sorted_by_sum[ptr].fi - 1] + sorted_by_sum[ptr].se >= z) {
auto [s, t] = sorted_by_sum[ptr++];
bit.upd(s, t, 1);
}
ans[ind[i]] = bit.query(x, sz(xs), y, INF + 1);
}
for (int i : rep(q)) cout << ans[i] << '\n';
}
int main()
{
cin.tie(0)->sync_with_stdio(0);
cin.exceptions(cin.failbit);
int testcase = 1;
// cin >> testcase;
while (testcase--)
{
solve();
}
#ifdef LOCAL
cerr << "Time elapsed: " << 1.0 * (double)clock() / CLOCKS_PER_SEC << " s.\n";
#endif
}
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
2 ms |
2820 KB |
Output is correct |
2 |
Correct |
1 ms |
2772 KB |
Output is correct |
3 |
Correct |
2 ms |
2772 KB |
Output is correct |
4 |
Correct |
1 ms |
2772 KB |
Output is correct |
5 |
Correct |
1 ms |
2768 KB |
Output is correct |
6 |
Correct |
2 ms |
2772 KB |
Output is correct |
7 |
Correct |
10 ms |
3412 KB |
Output is correct |
8 |
Correct |
10 ms |
3412 KB |
Output is correct |
9 |
Correct |
10 ms |
3284 KB |
Output is correct |
10 |
Correct |
6 ms |
3028 KB |
Output is correct |
11 |
Correct |
7 ms |
3412 KB |
Output is correct |
12 |
Incorrect |
5 ms |
2912 KB |
Output isn't correct |
13 |
Halted |
0 ms |
0 KB |
- |
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
378 ms |
17876 KB |
Output is correct |
2 |
Correct |
402 ms |
17972 KB |
Output is correct |
3 |
Correct |
387 ms |
17796 KB |
Output is correct |
4 |
Correct |
151 ms |
11656 KB |
Output is correct |
5 |
Correct |
247 ms |
18544 KB |
Output is correct |
6 |
Correct |
94 ms |
9092 KB |
Output is correct |
7 |
Correct |
312 ms |
17644 KB |
Output is correct |
8 |
Correct |
333 ms |
17704 KB |
Output is correct |
9 |
Correct |
296 ms |
17504 KB |
Output is correct |
10 |
Correct |
200 ms |
18668 KB |
Output is correct |
11 |
Correct |
122 ms |
9804 KB |
Output is correct |
12 |
Correct |
68 ms |
8708 KB |
Output is correct |
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
378 ms |
17876 KB |
Output is correct |
2 |
Correct |
402 ms |
17972 KB |
Output is correct |
3 |
Correct |
387 ms |
17796 KB |
Output is correct |
4 |
Correct |
151 ms |
11656 KB |
Output is correct |
5 |
Correct |
247 ms |
18544 KB |
Output is correct |
6 |
Correct |
94 ms |
9092 KB |
Output is correct |
7 |
Correct |
312 ms |
17644 KB |
Output is correct |
8 |
Correct |
333 ms |
17704 KB |
Output is correct |
9 |
Correct |
296 ms |
17504 KB |
Output is correct |
10 |
Correct |
200 ms |
18668 KB |
Output is correct |
11 |
Correct |
122 ms |
9804 KB |
Output is correct |
12 |
Correct |
68 ms |
8708 KB |
Output is correct |
13 |
Incorrect |
485 ms |
18288 KB |
Output isn't correct |
14 |
Halted |
0 ms |
0 KB |
- |
# |
Verdict |
Execution time |
Memory |
Grader output |
1 |
Correct |
2 ms |
2820 KB |
Output is correct |
2 |
Correct |
1 ms |
2772 KB |
Output is correct |
3 |
Correct |
2 ms |
2772 KB |
Output is correct |
4 |
Correct |
1 ms |
2772 KB |
Output is correct |
5 |
Correct |
1 ms |
2768 KB |
Output is correct |
6 |
Correct |
2 ms |
2772 KB |
Output is correct |
7 |
Correct |
10 ms |
3412 KB |
Output is correct |
8 |
Correct |
10 ms |
3412 KB |
Output is correct |
9 |
Correct |
10 ms |
3284 KB |
Output is correct |
10 |
Correct |
6 ms |
3028 KB |
Output is correct |
11 |
Correct |
7 ms |
3412 KB |
Output is correct |
12 |
Incorrect |
5 ms |
2912 KB |
Output isn't correct |
13 |
Halted |
0 ms |
0 KB |
- |