#include <bits/stdc++.h>
#include <atcoder/all>
#define rep(i,n) for(int i=0;i<n;i++)
#define reps(i,s,e) for(int i=s;i<e;i++)
#define rvrep(i,n) for(int i=n-1;i>=0;i--)
#define repv(i,v) for(auto i:v)
#define gcu getchar_unlocked
#define pcu putc_unlocked
#define all(x) x.begin(),x.end()
#define endln() pcu('\n',stdout)
#define space() pcu(' ',stdout)
template<typename T>inline bool chmax(T &a,T b){return((a<b)?(a=b,true):(false));}
template<typename T>inline bool chmin(T &a,T b){return((a>b)?(a=b,true):(false));}
using namespace std;
using namespace atcoder;
using ll=long long;
using vi=vector<int>;using vvi=vector<vi>;using vvvi=vector<vvi>;
using vl=vector<ll>;using vvl=vector<vl>;using vvvl=vector<vvl>;
using pii=pair<int,int>;using vpii=vector<pii>;using pll=pair<ll,ll>;using vpll=vector<pll>;using pli=pair<ll,int>;using pil=pair<int,ll>;
using mi=map<int,int>;using ms=map<string,int>;using mc=map<char,int>;using ml=map<ll,int>;
using si=set<int>;using sc=set<char>;using ss=set<string>;using sl=set<ll>;
using pi=pair<int,int>;using pl=pair<ll,ll>;
using mint=modint998244353;
using mint1=modint1000000007;

// input
int iin(){int r=0,s=1;int c=gcu();while(isspace(c))c=gcu();if(c=='-'){s=-1;c=gcu();}for(;!isspace(c);c=gcu())r=r*10+(c-'0');ungetc(c,stdin);return s*r;}
ll llin(){ll r=0;int s=1;int c=gcu();while(isspace(c))c=gcu();if(c=='-'){s=-1;c=gcu();}for(;!isspace(c);c=gcu())r=r*10+(c-'0');ungetc(c,stdin);return s*r;}
string input(){string r="";int c=gcu();while(isspace(c))c=gcu();for(;!isspace(c);c=gcu())r+=c;ungetc(c,stdin);return r;}
// output
template<typename T>void iout(T n){if(n<0){pcu('-',stdout);n=-n;}if(n<10){pcu('0'+n,stdout);pcu('\n',stdout);return;}int i;char b[21];b[20]=0;for(i=20;n>0;b[--i]='0'+n%10,n/=10);fputs(b+i,stdout);}
void print(string s){for(char i:s)pcu(i,stdout);}

class mComb {
 private:bool _calced = false;int _mod=1000000007;vector<long long>_fac,_invf,_inv;
 public:
  void setmod(int m){_mod=m;}
  void init(int n){_fac.resize(n+1);_invf.resize(n+1);_inv.resize(n+1);_fac[0]=1;_fac[1]=1;_invf[0]=1;_invf[1]=1;_inv[1]=1;reps(i,2,n+1){_fac[i]=_fac[i-1]*i%_mod;_inv[i]=_mod-_inv[_mod%i]*(_mod/i)%_mod;_invf[i]=_invf[i-1]*_inv[i]%_mod;}_calced=true;}
  long long F(int n){if(!_calced){cerr<<"not initialized\n";exit(1);}return _fac[n];}
  long long C(int n,int k){if(!_calced){cerr<<"not initialized\n";exit(1);}return _invf[k]*_invf[n-k]%_mod*_fac[n]%_mod;}
  long long P(int n,int k){if(!_calced){cerr<<"not initialized\n";exit(1);}return _invf[n-k]*_fac[n]%_mod;}
  long long H(int n,int k){if(!_calced){cerr<<"not initialized\n";exit(1);}return _invf[k]*_invf[n-1]%_mod*_fac[n+k-1]%_mod;}
};

int main() {
	ll n, p;
	cin >> n >> p;
	mComb c;
	c.setmod(998244353);
	c.init(500000);
	if (n < p) {
		cout << c.F(n) - 1 << '\n';
		return 0;
	}
	mint t = 0;
	for (int i = 0; i <= n / p; i++) {
		t += c.F(n) / (c.F(i) * c.F(n - i * p) * mint(p).pow(i));
	}
	cout << (c.F(n) - t).val() << '\n';
}