CF1661D WA

#include <assert.h>
#include <bits/stdc++.h>
using namespace std;
#define dbg(...) logger(#__VA_ARGS__, __VA_ARGS__)
template <typename... Args> void logger(string vars, Args &&... values)
{
    cerr << vars << " = ";
    string delim = "";
    (..., (cerr << delim << values, delim = ", "));
    cerr << endl;
}

template <class T> inline auto vv(int m) { return vector<vector<T>>(m, vector<T>(m)); }
template <class T> inline auto vv(int m, int n) { return vector<vector<T>>(m, vector<T>(n)); }
template <class T, T init> inline auto vv(int m) { return vector<vector<T>>(m, vector<T>(m, init)); }
template <class T, T init> inline auto vv(int m, int n) { return vector<vector<T>>(m, vector<T>(n, init)); }

template <class T> using mpq = priority_queue<T, vector<T>, greater<T>>;

using ll = long long;
using pii = pair<int, int>;
using vl = vector<ll>;
using vi = vector<int>;

int n, k;
vl b;

int main(int argc, char **argv)
{
    cin >> n >> k;
    b = vl(n);
    for (auto &x : b)
        cin >> x;
    ll op = 0;
    vl cnt(n);
    ll jcntj = 0;
    ll cntj = 0;
    for (int i = n - 1; i >= k - 1; --i) {
        ll val = b[i] - ((k + i) * cntj - jcntj);
        if (val > 0) {
            cnt[i] = (val + k - 1) / k;
            op += cnt[i];
        }
        // dbg(i, val, cnt[i], op);

        int knxt = i + k - 1;
        if (knxt < n) {
            jcntj = jcntj - knxt * cnt[knxt];
            cntj = cntj - cnt[knxt];
        }
        jcntj += i * cnt[i];
        cntj += cnt[i];
    }
    ll max_op = 0;
    for (int i = k - 2; i >= 0; --i) {
        ll val = b[i] - ((k + i) * cntj - jcntj);
        if (val > 0) {
            cnt[i] = (val + i) / (i + 1);
            max_op = max(max_op, cnt[i]);
        }
    }
    // dbg(op, max_op);
    cout << op + max_op << endl;
    return 0;
};