假设我们可以对每个位置快速维护一个数组,记录每个位置有哪些值是已经出现了的,哪些值是没有出现的,这样就可以决定修改的时候到底是 *2 还是 +1了。

但是很可惜,并不存在功能这么强大的数组,所以只能另寻方法啦。。。。

因为修改总是连续的一段的,所以我们可以发现,对于每个数值来说,被它覆盖的位置也是一段一段的,并且所有数的段数之和<=总的操作数(因为还会发生一些段的合并)。

所以我们可以对每个数维护一个端点集合set,每次修改的时候就暴力插入一段区间,看一看可以产生多少次合并。

这样的话,虽然一次修改的复杂度可能会十分之大,但是因为一个端点在贡献一次合并之后就被从 set 里删去了(一次合并只会对应线段树中的一次区间修改),并且每次最多增加两个端点,所以可以证明最后总的复杂度是 O( M * log (N) ),常数略大。。。

(并且用set维护每个数的区间的时候细节特别多,一定要想清楚了再去写)

(rank 22/212)

Discription

In the School of Magic in Dirtpolis a lot of interesting objects are studied on Computer Science lessons.

Consider, for example, the magic multiset. If you try to add an integer to it that is already presented in the multiset, each element in the multiset duplicates. For example, if you try to add the integer 22 to the multiset {1,2,3,3}{1,2,3,3}, you will get {1,1,2,2,3,3,3,3}{1,1,2,2,3,3,3,3}.

If you try to add an integer that is not presented in the multiset, it is simply added to it. For example, if you try to add the integer 44 to the multiset {1,2,3,3}{1,2,3,3}, you will get {1,2,3,3,4}{1,2,3,3,4}.

Also consider an array of nn initially empty magic multisets, enumerated from 11 to nn.

You are to answer qq queries of the form "add an integer xx to all multisets with indices l,l+1,…,rl,l+1,…,r" and "compute the sum of sizes of multisets with indices l,l+1,…,rl,l+1,…,r". The answers for the second type queries can be large, so print the answers modulo 998244353998244353.

Input

The first line contains two integers nn and qq (1≤n,q≤2⋅1051≤n,q≤2⋅105) — the number of magic multisets in the array and the number of queries, respectively.

The next qq lines describe queries, one per line. Each line starts with an integer tt(1≤t≤21≤t≤2) — the type of the query. If tt equals 11, it is followed by three integers ll, rr, xx (1≤l≤r≤n1≤l≤r≤n, 1≤x≤n1≤x≤n) meaning that you should add xx to all multisets with indices from ll to rr inclusive. If tt equals 22, it is followed by two integers ll, rr (1≤l≤r≤n1≤l≤r≤n) meaning that you should compute the sum of sizes of all multisets with indices from ll to rr inclusive.

Output

For each query of the second type print the sum of sizes of multisets on the given segment.

The answers can be large, so print them modulo 998244353998244353.

Examples

Input
4 4
1 1 2 1
1 1 2 2
1 1 4 1
2 1 4
Output
10
Input
3 7
1 1 1 3
1 1 1 3
1 1 1 2
1 1 1 1
2 1 1
1 1 1 2
2 1 1
Output
4
8

Note

In the first example after the first two queries the multisets are equal to [{1,2},{1,2},{},{}][{1,2},{1,2},{},{}], after the third query they are equal to [{1,1,2,2},{1,1,2,2},{1},{1}][{1,1,2,2},{1,1,2,2},{1},{1}].

In the second example the first multiset evolves as follows:

{}→{3}→{3,3}→{2,3,3}→{1,2,3,3}→{1,1,2,2,3,3,3,3}{}→{3}→{3,3}→{2,3,3}→{1,2,3,3}→{1,1,2,2,3,3,3,3}.

#include<bits/stdc++.h>
#define ll long long
using namespace std;
const int maxn=200005,ha=998244353;
#define lc (o<<1)
#define mid (l+r>>1)
#define rc ((o<<1)|1)
inline int add(int x,int y){ x+=y; return x>=ha?x-ha:x;}
inline void ADD(int &x,int y){ x+=y; if(x>=ha) x-=ha;}
struct node{
int p,id;
bool operator <(const node &u)const{
return p==u.p?id<u.id:p<u.p;
}
};
set<node> s[maxn];
set<node> :: iterator it,her;
int d[maxn*4],tag[maxn*4],sum[maxn*4];
int n,m,Q,le,ri,w,len[maxn*4],ans,opt; inline void maintain(int o){ sum[o]=add(sum[lc],sum[rc]);} inline void work(int o,int der){
ADD(tag[o],der);
ADD(sum[o],der*(ll)len[o]%ha);
} inline void mul(int o,int der){
d[o]=d[o]*(ll)der%ha;
tag[o]=tag[o]*(ll)der%ha;
sum[o]=sum[o]*(ll)der%ha;
} inline void pushdown(int o){
if(d[o]!=1){
mul(lc,d[o]),mul(rc,d[o]);
d[o]=1;
} if(tag[o]){
work(lc,tag[o]),work(rc,tag[o]);
tag[o]=0;
}
} void build(int o,int l,int r){
d[o]=1,len[o]=r-l+1;
if(l==r) return;
build(lc,l,mid);
build(rc,mid+1,r);
} void umul(int o,int l,int r){
if(l>=le&&r<=ri){ mul(o,2); return;}
pushdown(o);
if(le<=mid) umul(lc,l,mid);
if(ri>mid) umul(rc,mid+1,r);
maintain(o);
} void uadd(int o,int l,int r){
if(l>=le&&r<=ri){ work(o,w); return;}
pushdown(o);
if(le<=mid) uadd(lc,l,mid);
if(ri>mid) uadd(rc,mid+1,r);
maintain(o);
} void query(int o,int l,int r){
if(l>=le&&r<=ri){ ADD(ans,sum[o]); return;}
pushdown(o);
if(le<=mid) query(lc,l,mid);
if(ri>mid) query(rc,mid+1,r);
} void Change(int num,int L,int R){
set<node> &S=s[num]; int fl=L,fr=R; it=S.lower_bound((node){L,0}); for(w=1;L<=R;it=S.lower_bound((node){L,0})){
if(it==S.end()){
le=L,ri=R,uadd(1,1,n);
break;
}
else if(it->id){
le=L,ri=min(it->p,R),umul(1,1,n);
L=ri+1;
if(R>=it->p) S.erase(it);
}
else{
le=L,ri=min(it->p-1,R);
if(le<=ri) uadd(1,1,n);
L=ri+1;
if(R>=it->p) S.erase(it);
}
} it=S.lower_bound((node){fl,0});
her=S.lower_bound((node){fr,1}); if(it==S.begin()||(--it)->id) S.insert((node){fl,0});
if(her==S.end()||!her->id) S.insert((node){fr,1});
} inline void solve(){
build(1,1,n); while(Q--){
scanf("%d",&opt);
if(opt==1){
scanf("%d%d%d",&le,&ri,&w);
Change(w,le,ri);
}
else{
scanf("%d%d",&le,&ri),ans=0;
query(1,1,n),printf("%d\n",ans);
}
}
} int main(){
scanf("%d%d",&n,&Q);
solve();
return 0;
}

  

CodeForces - 981G Magic multisets的更多相关文章

  1. Codeforces 710C. Magic Odd Square n阶幻方

    C. Magic Odd Square time limit per test:1 second memory limit per test:256 megabytes input:standard ...

  2. Codeforces 670D1. Magic Powder - 1 暴力

    D1. Magic Powder - 1 time limit per test: 1 second memory limit per test: 256 megabytes input: stand ...

  3. [递推+矩阵快速幂]Codeforces 1117D - Magic Gems

    传送门:Educational Codeforces Round 60 – D   题意: 给定N,M(n <1e18,m <= 100) 一个magic gem可以分裂成M个普通的gem ...

  4. CodeForces 670D2 Magic Powder 二分

    D2. Magic Powder - 2 The term of this problem is the same as the previous one, the only exception — ...

  5. 【模拟】Codeforces 710C Magic Odd Square

    题目链接: http://codeforces.com/problemset/problem/710/C 题目大意: 构造一个N*N的幻方.任意可行解. 幻方就是每一行,每一列,两条对角线的和都相等. ...

  6. codeforces 632F. Magic Matrix

    题目链接 给一个n*n的矩阵, 问是否对角线上的元素全都为0, a[i][j]是否等于a[j][i], a[i][j]是否小于等于max(a[i][k], a[j][k]), k为任意值. 前两个都好 ...

  7. CodeForces - 710C Magic Odd Square(奇数和幻方构造)

    Magic Odd Square Find an n × n matrix with different numbers from 1 to n2, so the sum in each row, c ...

  8. Codeforces 632F Magic Matrix(bitset)

    题目链接  Magic Matrix 考虑第三个条件,如果不符合的话说明$a[i][k] < a[i][j]$ 或 $a[j][k] < a[i][j]$ 于是我们把所有的$(a[i][j ...

  9. codeforces C. Magic Formulas 解题报告

    题目链接:http://codeforces.com/problemset/problem/424/C 题目意思:给出 n 个数:p1, p2, ..., pn,定义: q1 = p1 ^ (1 mo ...

随机推荐

  1. 【BZOJ 4556】[Tjoi2016&Heoi2016]字符串 SAM+二分+主席树

    这道题市面上就两种法:一种是SA+二分+主席树,一种是SAM+二分+主席树(有不少人打线段树合并???)(除此之外还有一种利用炒鸡水的数据的暴力SA,贼快.....)(当时学SA的时候没做这道题,现在 ...

  2. 怎么替换jar包里面的文件?

    很多时候,我们需要替换包含在jar包里面的文件,例如修改里面的配置文件. 由于jar包已经生成,在不想重新用eclipse导出的情况下,我们怎么修改jar包里面的文件呢? 其实说出来很简单,可以使用以 ...

  3. HDU 5670

    Machine Time Limit: 2000/1000 MS (Java/Others)    Memory Limit: 65536/65536 K (Java/Others)Total Sub ...

  4. Codeforces Round #520 (Div. 2) C. Banh-mi

    C. Banh-mi time limit per test:1 second memory limit per test:256 megabytes 题目链接:https://codeforc.es ...

  5. HDFS集中化缓存管理

    概述 HDFS中的集中化缓存管理是一个明确的缓存机制,它允许用户指定要缓存的HDFS路径.NameNode会和保存着所需快数据的所有DataNode通信,并指导他们把块数据缓存在off-heap缓存中 ...

  6. 7月23号day15总结

    数据清洗完成之后开始编写前端,通过spring框架将清洗后数据库中的数据显示在页面中. 框架的搭建和js的使用都在学习阶段,

  7. Join an instance to my AWS Directory Service domain

    https://amazonaws-china.com/cn/premiumsupport/knowledge-center/ec2-systems-manager-dx-domain/ https: ...

  8. LCD实验学习笔记(四):系统时钟

    一般CPU频率(FCLK)高于内存.网卡等设备频率(HCLK),而串口.USB.I2C等设备频率(PCLK)更低. 系统时钟: 系统时钟源为晶振,初始频率12MHz. 通过设置MPLLCON寄存器的M ...

  9. 【Shell 编程基础第二部分】Shell里的流程控制、Shell里的函数及脚本调试方法!

    http://blog.csdn.net/xiaominghimi/article/details/7603003 本站文章均为李华明Himi原创,转载务必在明显处注明:转载自[黑米GameDev街区 ...

  10. gdb 调试 ncurses 全过程:

    转载地址: http://blog.jobbole.com/107759/ gdb 调试 ncurses 全过程: 发现网上的“gdb 示例”只有命令而没有对应的输出,我有点不满意.gdb 是 GNU ...