#include <bits/stdc++.h>
#pragma GCC optimize("Ofast")
using namespace std;
//#include "grader.cpp"
#include "peru.h"
typedef long long ll;
const int N=2500005,mod=(1e9)+7;
ll dp[N];
bitset<N>in_queue;
int solve(int n, int k, int* v){
    deque<int>dq;
    v--;
    priority_queue<pair<ll,pair<int,int>>,vector<pair<ll,pair<int,int>>>,greater<pair<ll,pair<int,int>>>>pq;
    for(int i=1;i<=n;i++){
        while(!dq.empty()&&dq.front()<i-k)in_queue[dq[0]]=0,dq.pop_front();
        while(!dq.empty()&&v[dq.back()]<=v[i])in_queue[dq.back()]=0,dq.pop_back();
        if(!dq.empty())pq.push({dp[dq.back()]+v[i],{dq.back(),i}});
        dq.push_back(i);
        in_queue[i]=1;
        while(!pq.empty()&&(!in_queue[pq.top().second.first]||!in_queue[pq.top().second.second]))pq.pop();
        dp[i]=LLONG_MAX;
        if(!pq.empty())dp[i]=pq.top().first;
        dp[i]=min(dp[i],dp[max(i-k,0)]+v[dq.front()]);
    }
    long long res=0;
    for(int i=1;i<=n;i++)res=(res*23+dp[i])%mod;
    return res;
}