[国家集训队]矩阵乘法

与静态kth的做法可以说是一模一样，计数改为用二维树状数组即可。

#include <bits/stdc++.h>
#define lowbit(x) x&(-x)
using namespace std;
int n,m,x,a,b,c,d,k,minn,maxn,cnt;
int sum[505][505],ans[60005];
struct node{int id1,id2,id3,id4,id,k,opt;}q[310005],q1[310005],q2[310005];

inline void change(int x,int y,int v)
{	
	int yy=y;
	while (x<=n)
	{
		y=yy;
		while (y<=n)
		{
			sum[x][y]+=v;
			y+=lowbit(y);
		}
		x+=lowbit(x);
	}
}
inline int query(int x,int y)
{
	int res=0;
	int yy=y;
	while (x)
	{
		y=yy;
		while (y)
		{
			res+=sum[x][y];
			y-=lowbit(y);
		}
		x-=lowbit(x);
	}
	return res;
}
inline int querysum(int a,int b,int c,int d)
{
	return query(c,d)-query(c,b-1)-query(a-1,d)+query(a-1,b-1);	
}

void solve(int ql,int qr,int l,int r)
{
	if (ql>qr) return;
	if (l==r)
	{
		for (register int i=ql; i<=qr; ++i) if (q[i].opt==2) ans[q[i].id]=l;
		return;	
	}
	int mid=l+r>>1;
	int p1=0,p2=0;
	for (register int i=ql; i<=qr; ++i)
	if (q[i].opt==1)
	{
		if (q[i].id1<=mid)
		{
			change(q[i].id3,q[i].id4,1);
			q1[++p1]=q[i];
		}
		else q2[++p2]=q[i];
	}
	else
	{
		int res=querysum(q[i].id1,q[i].id2,q[i].id3,q[i].id4);
		if (q[i].k<=res) q1[++p1]=q[i];
		else 
		{
			q[i].k-=res;
			q2[++p2]=q[i];
		}
	}
	for (register int i=1; i<=p1; ++i) if (q1[i].opt==1) change(q1[i].id3,q1[i].id4,-1);
	for (register int i=1; i<=p1; ++i) q[ql+i-1]=q1[i];
	for (register int i=1; i<=p2; ++i) q[ql+p1-1+i]=q2[i];
	solve(ql,ql+p1-1,l,mid);
	solve(ql+p1,qr,mid+1,r);
}

int main(){
	scanf("%d%d",&n,&m);
	minn=1e9; maxn=0;
	for (register int i=1; i<=n; ++i)
	for (register int j=1; j<=n; ++j)
	{
		scanf("%d",&x);
		minn=min(minn,x);
		maxn=max(maxn,x);
		q[++cnt]=(node){x,0,i,j,0,0,1};
	}
	for (register int i=1; i<=m; ++i)
	{
		scanf("%d%d%d%d%d",&a,&b,&c,&d,&k);
		q[++cnt]=(node){a,b,c,d,i,k,2};
	}
	solve(1,cnt,minn,maxn);
	for (register int i=1; i<=m; ++i) printf("%d\n",ans[i]);
return 0;	
}