test/yosupo_static_range_sum_with_upper_bound.test.cpp

View this file on GitHub
Last update: 2026-03-26 00:19:13+09:00
Problem: https://judge.yosupo.jp/problem/static_range_sum_with_upper_bound

Depends on

Code

#define PROBLEM "https://judge.yosupo.jp/problem/static_range_sum_with_upper_bound"

#include <algorithm>
#include <type_traits>
#include <vector>
using namespace std;

#include "../datastructure/weighted_wavelet_matrix.cpp"
#include <cstdio>
#include <cstring>
#include <string>

#include "../util/fastio.cpp"

int main() {
    Scanner in;
    Printer out;
    int n, q;
    in.read(n);
    in.read(q);
    vector<long long> a(n);
    for (int i = 0; i < n; ++i) in.read(a[i]);

    vector<long long> vals = a;
    sort(vals.begin(), vals.end());
    vals.erase(unique(vals.begin(), vals.end()), vals.end());
    vector<int> idx(n);
    for (int i = 0; i < n; ++i) idx[i] = (int)(lower_bound(vals.begin(), vals.end(), a[i]) - vals.begin());

    WeightedWaveletMatrix<long long, long long> wm;
    wm.build_from_index(idx, vals, a);
    while (q--) {
        int l, r;
        long long x;
        in.read(l);
        in.read(r);
        in.read(x);
        int xi = (int)(upper_bound(vals.begin(), vals.end(), x) - vals.begin());
        auto res = wm.count_sum_less_index(l, r, xi);
        out.println(res.count, res.sum);
    }
    return 0;
}

#line 1 "test/yosupo_static_range_sum_with_upper_bound.test.cpp"
#define PROBLEM "https://judge.yosupo.jp/problem/static_range_sum_with_upper_bound"

#include <algorithm>
#include <type_traits>
#include <vector>
using namespace std;

#line 1 "datastructure/weighted_wavelet_matrix.cpp"
template <class T, class U>
struct WeightedWaveletMatrix {
    struct CountSum {
        int count;
        U sum;
    };

    int n, lg, blocks;
    vector<int> mid;
    vector<unsigned long long> bit;
    vector<int> pref;
    vector<U> zero_sum;
    vector<U> base_sum;
    vector<T> vals;

    WeightedWaveletMatrix() : n(0), lg(0), blocks(0) {}
    WeightedWaveletMatrix(const vector<T> &v, const vector<U> &w) { build(v, w); }

    static inline void rank1_pair(const unsigned long long *row, const int *row_pref, int l, int r, int &l1, int &r1) {
        int l_block = l >> 6;
        l1 = row_pref[l_block];
        int l_rem = l & 63;
        if (l_rem) l1 += __builtin_popcountll(row[l_block] & ((1ULL << l_rem) - 1));

        int r_block = r >> 6;
        r1 = row_pref[r_block];
        int r_rem = r & 63;
        if (r_rem) r1 += __builtin_popcountll(row[r_block] & ((1ULL << r_rem) - 1));
    }

    template <class X>
    static auto encode_key(X x) -> typename make_unsigned<X>::type {
        using Key = typename make_unsigned<X>::type;
        Key key = static_cast<Key>(x);
        if constexpr (is_signed<X>::value) key ^= (Key(1) << (sizeof(X) * 8 - 1));
        return key;
    }

    void compress_generic(const vector<T> &v, vector<int> &cur) {
        vector<pair<T, int>> ord(n);
        for (int i = 0; i < n; ++i) ord[i] = {v[i], i};
        sort(ord.begin(), ord.end(), [](const pair<T, int> &a, const pair<T, int> &b) {
            return a.first < b.first;
        });
        vals.clear();
        vals.reserve(n);
        for (int i = 0; i < n; ++i) {
            if (vals.empty() || vals.back() < ord[i].first || ord[i].first < vals.back()) {
                vals.push_back(ord[i].first);
            }
            cur[ord[i].second] = (int)vals.size() - 1;
        }
    }

    void compress_integral(const vector<T> &v, vector<int> &cur) {
        using Key = typename make_unsigned<T>::type;
        vector<Key> keys(n);
        vector<int> ord(n), buf(n);
        for (int i = 0; i < n; ++i) {
            keys[i] = encode_key(v[i]);
            ord[i] = i;
        }

        const int B = 16;
        const int MASK = (1 << B) - 1;
        const int bucket_count = 1 << B;
        const int passes = (int)(sizeof(Key) * 8 + B - 1) / B;
        vector<int> cnt(bucket_count), pos(bucket_count);
        for (int pass = 0; pass < passes; ++pass) {
            fill(cnt.begin(), cnt.end(), 0);
            int shift = pass * B;
            for (int i = 0; i < n; ++i) ++cnt[(keys[ord[i]] >> shift) & MASK];
            pos[0] = 0;
            for (int i = 0; i + 1 < bucket_count; ++i) pos[i + 1] = pos[i] + cnt[i];
            for (int i = 0; i < n; ++i) {
                int id = ord[i];
                buf[pos[(keys[id] >> shift) & MASK]++] = id;
            }
            ord.swap(buf);
        }

        vals.clear();
        vals.reserve(n);
        bool has_prev = false;
        Key prev = 0;
        for (int i = 0; i < n; ++i) {
            int id = ord[i];
            if (!has_prev || keys[id] != prev) {
                vals.push_back(v[id]);
                prev = keys[id];
                has_prev = true;
            }
            cur[id] = (int)vals.size() - 1;
        }
    }

    void compress_values(const vector<T> &v, vector<int> &cur) {
        if constexpr (is_integral<T>::value && sizeof(T) <= 8) compress_integral(v, cur);
        else compress_generic(v, cur);
    }

    void build_from_index_internal(vector<int> cur, const vector<U> &w) {
        n = (int)cur.size();
        base_sum.assign(n + 1, U());
        for (int i = 0; i < n; ++i) base_sum[i + 1] = base_sum[i] + w[i];

        if (n == 0) {
            lg = 0;
            blocks = 0;
            mid.clear();
            bit.clear();
            pref.clear();
            zero_sum.clear();
            return;
        }

        int m = (int)vals.size();
        lg = 0;
        while ((1LL << lg) < m) ++lg;
        if (lg == 0) lg = 1;
        blocks = (n + 63) >> 6;
        vector<U> cur_w = w;

        mid.assign(lg, 0);
        bit.assign(lg * blocks, 0);
        pref.assign(lg * (blocks + 1), 0);
        zero_sum.assign(lg * (n + 1), U());
        vector<int> nxt(n);
        vector<U> nxt_w(n);

        for (int d = 0, shift = lg - 1; d < lg; ++d, --shift) {
            auto *row = bit.data() + d * blocks;
            auto *row_pref = pref.data() + d * (blocks + 1);
            auto *row_zero_sum = zero_sum.data() + d * (n + 1);
            int zero_cnt = 0;
            for (int i = 0; i < n; ++i) {
                int x = cur[i];
                int b = (x >> shift) & 1;
                if (b) row[i >> 6] |= 1ULL << (i & 63);
                else ++zero_cnt;
                row_zero_sum[i + 1] = row_zero_sum[i] + (b ? U() : cur_w[i]);
            }
            mid[d] = zero_cnt;
            for (int i = 0; i < blocks; ++i) row_pref[i + 1] = row_pref[i] + __builtin_popcountll(row[i]);

            int zi = 0, oi = zero_cnt;
            for (int i = 0; i < n; ++i) {
                int x = cur[i];
                if ((x >> shift) & 1) {
                    nxt[oi] = x;
                    nxt_w[oi++] = cur_w[i];
                }
                else {
                    nxt[zi] = x;
                    nxt_w[zi++] = cur_w[i];
                }
            }
            cur.swap(nxt);
            cur_w.swap(nxt_w);
        }
    }

    void build(const vector<T> &v, const vector<U> &w) {
        n = (int)v.size();
        if (n == 0) {
            lg = 0;
            blocks = 0;
            vals.clear();
            mid.clear();
            bit.clear();
            pref.clear();
            zero_sum.clear();
            base_sum.assign(1, U());
            return;
        }

        vector<int> cur(n);
        compress_values(v, cur);
        build_from_index_internal(move(cur), w);
    }

    void build_from_index(const vector<int> &idx, const vector<T> &sorted_vals, const vector<U> &w) {
        vals = sorted_vals;
        build_from_index_internal(idx, w);
    }

    CountSum count_sum_less_index(int l, int r, int xi) const {
        if (xi <= 0 || l >= r || n == 0) return {0, U()};
        if (xi >= (int)vals.size()) return {r - l, base_sum[r] - base_sum[l]};

        const int *mid_data = mid.data();
        const auto *bit_data = bit.data();
        const int *pref_data = pref.data();
        const U *zero_sum_data = zero_sum.data();
        CountSum res{0, U()};
        for (int d = 0, shift = lg - 1; d < lg; ++d, --shift) {
            int l1, r1;
            rank1_pair(bit_data, pref_data, l, r, l1, r1);
            int l0 = l - l1, r0 = r - r1;
            if ((xi >> shift) & 1) {
                res.count += r0 - l0;
                res.sum += zero_sum_data[r] - zero_sum_data[l];
                l = mid_data[d] + l1;
                r = mid_data[d] + r1;
            }
            else {
                l = l0;
                r = r0;
            }
            bit_data += blocks;
            pref_data += blocks + 1;
            zero_sum_data += n + 1;
        }
        return res;
    }

    CountSum count_sum_less(int l, int r, const T &x) const {
        int xi = (int)(lower_bound(vals.begin(), vals.end(), x) - vals.begin());
        return count_sum_less_index(l, r, xi);
    }

    CountSum count_sum_less_equal(int l, int r, const T &x) const {
        int xi = (int)(upper_bound(vals.begin(), vals.end(), x) - vals.begin());
        return count_sum_less_index(l, r, xi);
    }

    int count_less(int l, int r, const T &x) const {
        return count_sum_less(l, r, x).count;
    }

    int count_less_equal(int l, int r, const T &x) const {
        return count_sum_less_equal(l, r, x).count;
    }

    U sum_less(int l, int r, const T &x) const {
        return count_sum_less(l, r, x).sum;
    }

    U sum_less_equal(int l, int r, const T &x) const {
        return count_sum_less_equal(l, r, x).sum;
    }
};

/**
 * @brief 重み付きWavelet Matrix(Weighted Wavelet Matrix)
 */
#line 9 "test/yosupo_static_range_sum_with_upper_bound.test.cpp"
#include <cstdio>
#include <cstring>
#include <string>

#line 1 "util/fastio.cpp"
using namespace std;

extern "C" int fileno(FILE *);
extern "C" int isatty(int);

template<class T, class = void>
struct is_fastio_range : false_type {};

template<class T>
struct is_fastio_range<T, void_t<decltype(declval<T &>().begin()), decltype(declval<T &>().end())>> : true_type {};

template<class T, class = void>
struct has_fastio_value : false_type {};

template<class T>
struct has_fastio_value<T, void_t<decltype(declval<const T &>().value())>> : true_type {};

struct FastIoDigitTable {
    char num[40000];

    constexpr FastIoDigitTable() : num() {
        for (int i = 0; i < 10000; ++i) {
            int x = i;
            for (int j = 3; j >= 0; --j) {
                num[i * 4 + j] = char('0' + x % 10);
                x /= 10;
            }
        }
    }
};

struct Scanner {
    static constexpr int BUFSIZE = 1 << 17;
    static constexpr int OFFSET = 64;
    char buf[BUFSIZE + 1];
    int idx, size;
    bool interactive;

    Scanner() : idx(0), size(0), interactive(isatty(fileno(stdin))) {}

    inline void load() {
        int len = size - idx;
        memmove(buf, buf + idx, len);
        if (interactive) {
            if (fgets(buf + len, BUFSIZE + 1 - len, stdin)) size = len + (int)strlen(buf + len);
            else size = len;
        } else {
            size = len + (int)fread(buf + len, 1, BUFSIZE - len, stdin);
        }
        idx = 0;
        buf[size] = 0;
    }

    inline void ensure() {
        if (idx + OFFSET > size) load();
    }

    inline void ensure_interactive() {
        if (idx == size) load();
    }

    inline char skip() {
        if (interactive) {
            ensure_interactive();
            while (buf[idx] && buf[idx] <= ' ') {
                ++idx;
                ensure_interactive();
            }
            return buf[idx++];
        }
        ensure();
        while (buf[idx] && buf[idx] <= ' ') {
            ++idx;
            ensure();
        }
        return buf[idx++];
    }

    template<class T, typename enable_if<is_integral<T>::value, int>::type = 0>
    void read(T &x) {
        if (interactive) {
            char c = skip();
            bool neg = false;
            if constexpr (is_signed<T>::value) {
                if (c == '-') {
                    neg = true;
                    ensure_interactive();
                    c = buf[idx++];
                }
            }
            x = 0;
            while (c >= '0') {
                x = x * 10 + (c & 15);
                ensure_interactive();
                c = buf[idx++];
            }
            if constexpr (is_signed<T>::value) {
                if (neg) x = -x;
            }
            return;
        }
        char c = skip();
        bool neg = false;
        if constexpr (is_signed<T>::value) {
            if (c == '-') {
                neg = true;
                c = buf[idx++];
            }
        }
        x = 0;
        while (c >= '0') {
            x = x * 10 + (c & 15);
            c = buf[idx++];
        }
        if constexpr (is_signed<T>::value) {
            if (neg) x = -x;
        }
    }

    template<class T, typename enable_if<!is_integral<T>::value && !is_fastio_range<T>::value && !is_same<typename decay<T>::type, string>::value && has_fastio_value<T>::value, int>::type = 0>
    void read(T &x) {
        long long v;
        read(v);
        x = T(v);
    }

    template<class Head, class Next, class... Tail>
    void read(Head &head, Next &next, Tail &...tail) {
        read(head);
        read(next, tail...);
    }

    template<class T, class U>
    void read(pair<T, U> &p) {
        read(p.first, p.second);
    }

    template<class T, typename enable_if<is_fastio_range<T>::value && !is_same<typename decay<T>::type, string>::value, int>::type = 0>
    void read(T &a) {
        for (auto &x : a) read(x);
    }

    void read(char &c) {
        c = skip();
    }

    void read(string &s) {
        s.clear();
        if (interactive) {
            ensure_interactive();
            while (buf[idx] && buf[idx] <= ' ') {
                ++idx;
                ensure_interactive();
            }
            while (true) {
                int start = idx;
                while (idx < size && buf[idx] > ' ') ++idx;
                s.append(buf + start, idx - start);
                if (idx < size) break;
                load();
                if (size == 0) break;
            }
            if (idx < size) ++idx;
            return;
        }
        ensure();
        while (buf[idx] && buf[idx] <= ' ') {
            ++idx;
            ensure();
        }
        while (true) {
            int start = idx;
            while (idx < size && buf[idx] > ' ') ++idx;
            s.append(buf + start, idx - start);
            if (idx < size) break;
            load();
        }
        if (idx < size) ++idx;
    }
};

struct Printer {
    static constexpr int BUFSIZE = 1 << 17;
    static constexpr int OFFSET = 64;
    char buf[BUFSIZE];
    int idx;
    bool interactive;
    inline static constexpr FastIoDigitTable table{};

    Printer() : idx(0), interactive(isatty(fileno(stdout))) {}
    ~Printer() { flush(); }

    inline void flush() {
        if (idx) {
            fwrite(buf, 1, idx, stdout);
            idx = 0;
        }
    }

    inline void pc(char c) {
        if (idx > BUFSIZE - OFFSET) flush();
        buf[idx++] = c;
        if (interactive && c == '\n') flush();
    }

    inline void print_range(const char *s, size_t n) {
        size_t pos = 0;
        while (pos < n) {
            if (idx == BUFSIZE) flush();
            size_t chunk = min(n - pos, (size_t)(BUFSIZE - idx));
            memcpy(buf + idx, s + pos, chunk);
            idx += (int)chunk;
            pos += chunk;
        }
    }

    void print(const char *s) {
        print_range(s, strlen(s));
    }

    void print(const string &s) {
        print_range(s.data(), s.size());
    }

    void print(char c) {
        pc(c);
    }

    void print(bool b) {
        pc(char('0' + (b ? 1 : 0)));
    }

    template<class T, typename enable_if<is_integral<T>::value && !is_same<T, bool>::value, int>::type = 0>
    void print(T x) {
        if (idx > BUFSIZE - 100) flush();
        using U = typename make_unsigned<T>::type;
        U y;
        if constexpr (is_signed<T>::value) {
            if (x < 0) {
                buf[idx++] = '-';
                y = U(0) - static_cast<U>(x);
            } else {
                y = static_cast<U>(x);
            }
        } else {
            y = x;
        }
        if (y == 0) {
            buf[idx++] = '0';
            return;
        }
        static constexpr int TMP_SIZE = sizeof(U) * 10 / 4;
        char tmp[TMP_SIZE];
        int pos = TMP_SIZE;
        while (y >= 10000) {
            pos -= 4;
            memcpy(tmp + pos, table.num + (y % 10000) * 4, 4);
            y /= 10000;
        }
        if (y >= 1000) {
            memcpy(buf + idx, table.num + (y << 2), 4);
            idx += 4;
        } else if (y >= 100) {
            memcpy(buf + idx, table.num + (y << 2) + 1, 3);
            idx += 3;
        } else if (y >= 10) {
            unsigned q = (unsigned(y) * 205) >> 11;
            buf[idx] = char('0' + q);
            buf[idx + 1] = char('0' + (unsigned(y) - q * 10));
            idx += 2;
        } else {
            buf[idx++] = char('0' + y);
        }
        memcpy(buf + idx, tmp + pos, TMP_SIZE - pos);
        idx += TMP_SIZE - pos;
    }

    template<class T, typename enable_if<!is_integral<T>::value && !is_fastio_range<T>::value && !is_same<typename decay<T>::type, string>::value && has_fastio_value<T>::value, int>::type = 0>
    void print(const T &x) {
        print(x.value());
    }

    template<class T, typename enable_if<is_fastio_range<T>::value && !is_same<typename decay<T>::type, string>::value, int>::type = 0>
    void print(const T &a) {
        bool first = true;
        for (auto &&x : a) {
            if (!first) pc(' ');
            first = false;
            print(x);
        }
    }

    template<class T>
    void println(const T &x) {
        print(x);
        pc('\n');
    }

    template<class Head, class... Tail>
    void println(const Head &head, const Tail &...tail) {
        print(head);
        ((pc(' '), print(tail)), ...);
        pc('\n');
    }

    void println() {
        pc('\n');
    }
};

template<class T>
Scanner &operator>>(Scanner &in, T &x) {
    in.read(x);
    return in;
}

template<class T>
Printer &operator<<(Printer &out, const T &x) {
    out.print(x);
    return out;
}

/**
 * @brief 高速入出力(Fast IO)
 */
#line 14 "test/yosupo_static_range_sum_with_upper_bound.test.cpp"

int main() {
    Scanner in;
    Printer out;
    int n, q;
    in.read(n);
    in.read(q);
    vector<long long> a(n);
    for (int i = 0; i < n; ++i) in.read(a[i]);

    vector<long long> vals = a;
    sort(vals.begin(), vals.end());
    vals.erase(unique(vals.begin(), vals.end()), vals.end());
    vector<int> idx(n);
    for (int i = 0; i < n; ++i) idx[i] = (int)(lower_bound(vals.begin(), vals.end(), a[i]) - vals.begin());

    WeightedWaveletMatrix<long long, long long> wm;
    wm.build_from_index(idx, vals, a);
    while (q--) {
        int l, r;
        long long x;
        in.read(l);
        in.read(r);
        in.read(x);
        int xi = (int)(upper_bound(vals.begin(), vals.end(), x) - vals.begin());
        auto res = wm.count_sum_less_index(l, r, xi);
        out.println(res.count, res.sum);
    }
    return 0;
}