#include <bits/allocator.h>

#pragma GCC optimize("Ofast")
#pragma GCC target("avx2")

#define

// cf bits/extc++.h nonsense
#ifdef ONLINE_JUDGE
#define _EXT_CODECVT_SPECIALIZATIONS_H 1
#define _EXT_ENC_FILEBUF_H 1
#endif
#include "bits/extc++.h"

using namespace std;

template <typename T, typename... U>
void dbgh(const T& t, const U&... u) {
    cerr << t;
    ((cerr << " | " << u), ...);
    cerr << endl;
}

#ifdef DEBUG
#define dbg(...)                                              \
    cerr << "L" << __LINE__ << " [" << #__VA_ARGS__ << "]: "; \
    dbgh(__VA_ARGS__)
#else
#define dbg(...)
#define cerr   \
    if (false) \
    cerr
#endif

using ll = long long;

#define endl "\n"
#define long int64_t
#define sz(x) int(std::size(x))

inline void init_io() {
    cin.tie(nullptr);
    cin.exceptions(ios::failbit);
    ios_base::sync_with_stdio(false);
}

template <typename T>
ostream& operator<<(ostream& out, const vector<T>& arr) {
    out << "[";
    for (int i = 0; i < sz(arr); i++) {
        if (i) {
            out << ", ";
        }
        out << arr[i];
    }
    return out << "]";
}

template <typename T>
struct PSA {
    int n;
    vector<T> psum;

    template <typename U>
    PSA(const vector<U>& arr) : n(sz(arr)), psum(n + 1) {
        for (int i = 0; i < n; i++) {
            psum[i + 1] = psum[i] + arr[i];
        }
    }

    T query(int l, int r) const {
        return psum[r] - psum[l];
    }
};

template <typename T>
bool on(T mask, int bit) {
    return (mask >> bit) & 1;
}

void solve() {
    int n, kv;
    cin >> n >> kv;

    vector<long> costs(n), vals(n);
    for (auto& a : costs) {
        cin >> a;
    }
    for (auto& a : vals) {
        cin >> a;
    }

    PSA<long> costs_psa(costs);

    long buf1[kv + 1] alignas(1024), buf2[kv + 1] alignas(1024);

    long *cdp alignas(1024) = buf1, *pdp alignas(1024) = buf2;
    fill_n(cdp, kv + 1, -1e18);
    fill_n(pdp, kv + 1, -1e18);
    // vector<long> cdp(kv + 1, long(-1e18)), pdp(kv + 1, long(-1e18));
    cdp[kv] = -costs_psa.psum[n];
    // vector dp(n + 1, vector(kv + 1, long(-1e18)));
    // for (int i = 0; i <= n; i++) {
    // dp[i][kv] = -costs_psa.psum[i];
    // }

    long ans = -1e18;

    for (int i = n - 1; i >= 0; i--) {
        swap(cdp, pdp);

        const int mx_iter = min(kv, (i + 16) & (~15));
        // const int mx_iter = kv;

#pragma GCC ivdep
        for (int j = max(0, (kv - (n - i) - 10) & (~15)); j < mx_iter; j++) {
            cdp[j] = max(pdp[j], pdp[j + 1] + vals[i]);
        }

        ans = max(ans, cdp[0] + costs_psa.psum[i]);
        cdp[kv] = -costs_psa.psum[i];
    }

    // for (int i = 0; i <= n; i++) {
    // ans = max(ans, dp[i][0] + costs_psa.psum[i]);
    // }

    cout << ans << endl;
}

int main() {
    init_io();
    solve();
}

answer.code:6:8: error: no macro name given in #define directive
    6 | #define
      |        ^