#include <bits/stdc++.h>
using namespace std;
using ll = long long;
using db = long double; // or double, if TL is tight
using str = string; // yay python!
// pairs
using pii = pair<ll,int>;
using pl = pair<ll,ll>;
using pd = pair<db,db>;
#define mp make_pair
#define f first
#define s second
#define tcT template<class T
#define tcTU tcT, class U
// ^ lol this makes everything look weird but I'll try it
tcT> using V = vector<T>;
tcT, size_t SZ> using AR = array<T,SZ>;
using vi = V<int>;
using vb = V<bool>;
using vd = V<db>;
using vs = V<str>;
using vpi = V<pii>;
using vpl = V<pl>;
using vpd = V<pd>;
// vectors
// oops size(x), rbegin(x), rend(x) need C++17
#define sz(x) int((x).size())
#define bg(x) begin(x)
#define all(x) bg(x), end(x)
#define rall(x) x.rbegin(), x.rend()
#define sor(x) sort(all(x))
#define rsz resize
#define ins insert
#define pb push_back
#define eb emplace_back
#define ft front()
#define bk back()
#define lb lower_bound
#define ub upper_bound
#define FOR(i,a,b) for (int i = (a); i < (b); ++i)
#define F0R(i,a) FOR(i,0,a)
#define ROF(i,a,b) for (int i = (b)-1; i >= (a); --i)
#define R0F(i,a) ROF(i,0,a)
#define rep(a) F0R(_,a)
#define each(a,x) for (auto& a: x)
const int MOD = 998244353;
const int N = 3e5 + 5;
const ll inf = 1e15; // not too close to LLONG_MAX
const db PI = acos((db)-1);
const int dx[4]{1,0,-1,0}, dy[4]{0,1,0,-1}; // for every grid problem!!
mt19937 rng((uint32_t)chrono::steady_clock::now().time_since_epoch().count());
template<class T> using pqg = priority_queue<T,vector<T>,greater<T>>;
struct DSU {
vi e; void init(int N) { e = vi(N,-1); }
int get(int x) { return e[x] < 0 ? x : e[x] = get(e[x]); }
bool sameSet(int a, int b) { return get(a) == get(b); }
int size(int x) { return -e[get(x)]; }
bool unite(int x, int y) { // union by size
x = get(x), y = get(y); if (x == y) return 0;
if (e[x] > e[y]) swap(x,y);
e[x] += e[y]; e[y] = x; return 1;
}
};
/*
inline namespace Helpers {
//////////// is_iterable
// https://stackoverflow.com/questions/13830158/check-if-a-variable-type-is-iterable
// this gets used only when we can call begin() and end() on that type
tcT, class = void> struct is_iterable : false_type {};
tcT> struct is_iterable<T, void_t<decltype(begin(declval<T>())),
decltype(end(declval<T>()))
>
> : true_type {};
tcT> constexpr bool is_iterableftchv = is_iterable<T>::value;
//////////// is_readable
tcT, class = void> struct is_readable : false_type {};
tcT> struct is_readable<T,
typename std::enableftchif_t<
is_same_v<decltype(cin >> declval<T&>()), istream&>
>
> : true_type {};
tcT> constexpr bool is_readableftchv = is_readable<T>::value;
//////////// is_printable
// // https://nafe.es/posts/2020-02-29-is-printable/
tcT, class = void> struct is_printable : false_type {};
tcT> struct is_printable<T,
typename std::enableftchif_t<
is_same_v<decltype(cout << declval<T>()), ostream&>
>
> : true_type {};
tcT> constexpr bool is_printableftchv = is_printable<T>::value;
}*/
using ll = long long;
using db = long double; // or double, if TL is tight
using str = string; // yay python!
// pairs
using pii = pair<ll,int>;
using pl = pair<ll,ll>;
using pd = pair<db,db>;
#define mp make_pair
#define f first
#define s second
#define tcT template<class T
#define tcTU tcT, class U
// ^ lol this makes everything look weird but I'll try it
tcT> using V = vector<T>;
tcT, size_t SZ> using AR = array<T,SZ>;
using vi = V<int>;
using vb = V<bool>;
using vd = V<db>;
using vs = V<str>;
using vpi = V<pii>;
using vpl = V<pl>;
using vpd = V<pd>;
// vectors
// oops size(x), rbegin(x), rend(x) need C++17
#define sz(x) int((x).size())
#define bg(x) begin(x)
#define all(x) bg(x), end(x)
#define rall(x) x.rbegin(), x.rend()
#define sor(x) sort(all(x))
#define rsz resize
#define ins insert
#define pb push_back
#define eb emplace_back
#define ft front()
#define bk back()
#define lb lower_bound
#define ub upper_bound
#define FOR(i,a,b) for (int i = (a); i < (b); ++i)
#define F0R(i,a) FOR(i,0,a)
#define ROF(i,a,b) for (int i = (b)-1; i >= (a); --i)
#define R0F(i,a) ROF(i,0,a)
#define rep(a) F0R(_,a)
#define each(a,x) for (auto& a: x)
template<class T> using pqg = priority_queue<T,vector<T>,greater<T>>;
#define int long long
/*
inline namespace Helpers {
//////////// is_iterable
// https://stackoverflow.com/questions/13830158/check-if-a-variable-type-is-iterable
// this gets used only when we can call begin() and end() on that type
tcT, class = void> struct is_iterable : false_type {};
tcT> struct is_iterable<T, void_t<decltype(begin(declval<T>())),
decltype(end(declval<T>()))
>
> : true_type {};
tcT> constexpr bool is_iterableftchv = is_iterable<T>::value;
//////////// is_readable
tcT, class = void> struct is_readable : false_type {};
tcT> struct is_readable<T,
typename std::enableftchif_t<
is_same_v<decltype(cin >> declval<T&>()), istream&>
>
> : true_type {};
tcT> constexpr bool is_readableftchv = is_readable<T>::value;
//////////// is_printable
// // https://nafe.es/posts/2020-02-29-is-printable/
tcT, class = void> struct is_printable : false_type {};
tcT> struct is_printable<T,
typename std::enableftchif_t<
is_same_v<decltype(cout << declval<T>()), ostream&>
>
> : true_type {};
tcT> constexpr bool is_printableftchv = is_printable<T>::value;
}*/
const int M = 4000000;
int t, n, m, ans = -inf, leftch[M], rigch[M], cur, sz, rt[N];
map<int, int> mpp;
vector<int> v;
pii a[N];
struct nod
{
int sum, cnt;
};
nod tree[M];
nod merge(nod a, nod b)
{
return {
a.sum + b.sum, a.cnt + b.cnt
};
}
void build(int vert, int le, int ri)
{
if (le == ri)
{
return;
}
int mid = (le + ri) / 2;
leftch[vert] = ++cur;
rigch[vert] = ++cur;
build(leftch[vert], le, mid);
build(rigch[vert], mid + 1, ri);
tree[vert] = merge(tree[leftch[vert]], tree[rigch[vert]]);
}
void update(int vert, int le, int ri, int idx, int val)
{
if (le > idx || ri < idx) return;
if (le == ri)
{
tree[cur].sum = tree[vert].sum + val;
tree[cur].cnt = tree[vert].cnt + 1;
return;
}
int x = cur;
int mid = (le + ri) / 2;
leftch[x] = leftch[vert];
rigch[x] = rigch[vert];
if (idx <= mid) leftch[x] = ++cur;
else rigch[x] = ++cur;
update(leftch[vert], le, mid, idx, val);
update(rigch[vert], mid + 1, ri, idx, val);
tree[x] = merge(tree[leftch[x]], tree[rigch[x]]);
}
int answer(int le, int ri, int vert1, int vert2, int k)
{
if (le > ri) return 0;
if (le == ri)
{
int occ = tree[vert2].cnt - tree[vert1].cnt;
int vl = (tree[vert2].sum - tree[vert1].sum) / occ;
return k * vl;
}
int mid = (le + ri) / 2;
int raod = tree[rigch[vert2]].cnt - tree[rigch[vert1]].cnt;
if (raod >= k)
{
return answer(mid + 1, ri, rigch[vert1], rigch[vert2], k);
}
else
{
return tree[rigch[vert2]].sum - tree[rigch[vert1]].sum + answer(le, mid, leftch[vert1], leftch[vert2], k - raod);
}
}
int cost(int le, int ri)
{
return answer(1, sz, rt[le - 1], rt[ri], m - 2);
}
vector<int> comp(vector<int> v)
{
sort(v.begin(), v.end());
vector<int> res;
for (int i = 0; i < v.size(); i++)
{
if (i == 0 || v[i] != v[i - 1]) res.pb(v[i]);
}
return res;
}
void solve(int le, int ri, int optl, int optr)
{
int mid = (le + ri) / 2;
int best = -inf, bestid = 0, cursum = 0;
multiset<int> s;
for (int i = max(mid + m - 1, optl); i <= optr; i++)
{
int x = -2 *a[i].f + cost(mid + 1, i - 1) + a[i].s + a[mid].s;
if (x > best) best = x, bestid = i;
}
ans = max(ans, best + 2 *a[mid].f);
if (le == ri) return;
solve(le, mid, optl, bestid);
solve(mid + 1, ri, bestid, optr);
}
signed main()
{
ios_base::sync_with_stdio(false), cin.tie(NULL), cout.tie(NULL);
cin >> n >> m;
for (int i = 1; i <= n; i++)
{
cin >> a[i].s >> a[i].f;
v.pb(a[i].s);
}
sort(a + 1, a + n + 1);
vector<int> v1 = comp(v);
v.clear();
v.pb(0);
for (int x: v1) v.pb(x);
sz = v.size() - 1;
for (int i = 1; i <= sz; i++)
{
mpp[v[i]] = i;
}
cur = 1;
rt[0] = 1;
build(1, 1, sz);
for (int i = 1; i <= n; i++)
{
int prt = rt[i - 1];
rt[i] = ++cur;
update(rt[i - 1], 1, sz, mpp[a[i].s], a[i].s);
}
solve(1, n - m + 1, 1, n);
cout << ans << "\n";
}
