[分块] BZOJ 4867 [Ynoi2017]舌尖上的由乃

这个题啊 分块 每块$O(\sqrt n)$
把每个块内出现次数做一个前缀和，方便二分的时候O(1)查
然后整块加就打标记，两边零散的，变化不超过10，那么就直接在前缀和数组上修改下
复杂度大概是$O(n\sqrt n\log n+n\sqrt n\times 10)$
还有个问题是前缀和存不下？其实是存的下的，我们只要求从最小值开到最大值就好了，每次整块加不会改变极差，散块加每次只会加一个10

#include<cstdio>
#include<cstdlib>
#include<algorithm>
#include<cmath>
using namespace std;

inline char nc(){
  static char buf[100000],*p1=buf,*p2=buf;
  return p1==p2&&(p2=(p1=buf)+fread(buf,1,100000,stdin),p1==p2)?EOF:*p1++;
}
inline void read(int &x){
  char c=nc(),b=1;
  for (;!(c>='0' && c<='9');c=nc()) if (c=='-') b=-1;
  for (x=0;c>='0' && c<='9';x=x*10+c-'0',c=nc()); x*=b;
}

const int N=100005;

namespace T{
  struct edge{
    int u,v,w,next;
  }G[N<<1];
  int head[N],inum;
#define V G[p].v
  inline void add(int u,int v,int w,int p){
    G[p].u=u; G[p].v=v; G[p].w=w; G[p].next=head[u]; head[u]=p;
  }
  int clk,pre[N],size[N],dis[N];
  inline void dfs(int u){
    pre[u]=++clk; size[u]=1;
    for (int p=head[u];p;p=G[p].next)
      dis[V]=dis[u]+G[p].w,dfs(V),size[u]+=size[V];
  }
}
using namespace T;

const int BB=505;
const int M=20000005;

int n,Q,len,_a[N];
int minv[BB],maxv[BB],Size[BB];
int B,pos[N],cnt,lp[BB],rp[BB];

int tag[BB],F[M];

#define f(i,j) (*(F+Size[(i)-1]+(j)-minv[(i)]))

int qo[N],ql[N],qr[N],qk[N];

inline void Build(){
  B=sqrt(n);
  for (int i=1;i<=n;i++) pos[i]=(i-1)/B+1; cnt=pos[n];
  for (int i=1;i<=cnt;i++) lp[i]=(i-1)*B+1,rp[i]=i*B; rp[cnt]=n;
  for (int i=1;i<=cnt;i++) minv[i]=1<<30,maxv[i]=-1<<30;
  for (int i=1;i<=n;i++) minv[pos[i]]=min(minv[pos[i]],_a[i]),maxv[pos[i]]=max(maxv[pos[i]],_a[i]);
  for (int i=1;i<=Q;i++) if (qo[i]==2) maxv[pos[ql[i]]]+=qk[i],maxv[pos[qr[i]]]+=qk[i];
  for (int i=1;i<=cnt;i++) minv[i]--,Size[i]=maxv[i]-minv[i]+Size[i-1];
  for (int i=1;i<=n;i++) f(pos[i],_a[i])++;
  for (int i=1;i<=cnt;i++)
    for (int j=minv[i]+2;j<=maxv[i];j++)
      f(i,j)+=f(i,j-1);
}

inline void Add(int l,int r,int w){
  int lb=pos[l],rb=pos[r];
  if (lb==rb){
    for (int i=l;i<=r;i++){
      for (int j=_a[i]+tag[lb];j<=_a[i]+tag[lb]+w-1;j++)
    f(lb,j)--;
      _a[i]+=w;
    }
    return;
  }
  for (int i=lb+1;i<rb;i++) tag[i]+=w,minv[i]+=w;
  for (int i=l;i<=rp[lb];i++){
    for (int j=_a[i]+tag[lb];j<=_a[i]+tag[lb]+w-1;j++)
      f(lb,j)--;
    _a[i]+=w;
  }
  for (int i=lp[rb];i<=r;i++){
    for (int j=_a[i]+tag[rb];j<=_a[i]+tag[rb]+w-1;j++)
      f(rb,j)--;
    _a[i]+=w;
  }
}

int tmp[N],pnt;

inline int Query(int l,int r,int K){
  if (K>r-l+1) return -1;
  int lb=pos[l],rb=pos[r];
  if (lb==rb){
    pnt=0;
    for (int i=l;i<=r;i++) tmp[++pnt]=_a[i]+tag[lb];
    nth_element(tmp+1,tmp+K,tmp+pnt+1);
    return tmp[K];
  }
  pnt=0;
  for (int i=l;i<=rp[lb];i++) tmp[++pnt]=_a[i]+tag[lb];
  for (int i=lp[rb];i<=r;i++) tmp[++pnt]=_a[i]+tag[rb];
  sort(tmp+1,tmp+pnt+1);
  int L=-1,R=(n+Q)*len;
  while (L+1<R){
    int MID=(L+R)>>1,tot=0;
    if (L<100000 && R>100000) MID=100000;
    for (int i=lb+1;i<rb;i++) if (MID>minv[i]) tot+=f(i,min(MID,maxv[i]+tag[i]));
    tot+=upper_bound(tmp+1,tmp+pnt+1,MID)-tmp-1;
    if (tot<K)
      L=MID;
    else
      R=MID;
  }
  return R;
}

int main(){
  int x,y;
  freopen("j.in","r",stdin);
  freopen("j.out","w",stdout);
  read(n); read(Q); read(len);
  for (int i=2;i<=n;i++) read(x),read(y),add(x,i,y,++inum);
  dfs(1);
  for (int i=1;i<=n;i++) _a[pre[i]]=dis[i];
  for (int i=1;i<=Q;i++) read(qo[i]),read(x),read(qk[i]),ql[i]=pre[x],qr[i]=pre[x]+size[x]-1;
  Build();
  for (int i=1;i<=Q;i++)
    if (qo[i]==1){
      int ans=Query(ql[i],qr[i],qk[i]);
      printf("%d\n",ans);
    }else
      Add(ql[i],qr[i],qk[i]);
  return 0;
}

还有一种理论复杂度更优的算法
每块$O(\sqrt n\log n)$
散块加的时候我们归并，那么复杂度就是$O(n\sqrt n\log n)$
二分，因为块的大小是$O(\sqrt n\log n)$，那么复杂度算一下还是$O(n\sqrt n\log n)$
然而归并、二分的常数爆炸

#include<cstdio>
#include<cstdlib>
#include<cmath>
#include<algorithm>
using namespace std;
typedef pair<int,int> abcd;

inline char nc(){
  static char buf[100000],*p1=buf,*p2=buf;
  return p1==p2&&(p2=(p1=buf)+fread(buf,1,100000,stdin),p1==p2)?EOF:*p1++;
}
inline void read(int &x){
  char c=nc(),b=1;
  for (;!(c>='0' && c<='9');c=nc()) if (c=='-') b=-1;
  for (x=0;c>='0' && c<='9';x=x*10+c-'0',c=nc()); x*=b;
}

namespace T{
  const int N=100005;
  struct edge{
    int u,v,w,next;
  }G[N<<1];
  int head[N],inum;
  inline void add(int u,int v,int w,int p){
    G[p].u=u; G[p].v=v; G[p].w=w; G[p].next=head[u]; head[u]=p;
  }
#define V G[p].v
  int pre[N],clk,size[N],dis[N];
  inline void dfs(int u){
    pre[u]=++clk; size[u]=1;
    for (int p=head[u];p;p=G[p].next)
      dis[V]=dis[u]+G[p].w,dfs(V),size[u]+=size[V];
  }
}
using namespace T;

const int BB=5605;
const int NB=505;

int C0,C1,C2;

int n,Q,len;
int _a[N];
inline bool cmp(int x,int y){
  return _a[x]<_a[y];
}
int B,cnt,lp[NB],rp[NB];
int pos[N];
int a[NB][BB],tag[NB];

inline void Build(){
  B=min(n,(int)(sqrt(n)*5));
  for (int i=1;i<=n;i++) pos[i]=(i-1)/B+1; cnt=pos[n];
  for (int i=1;i<=cnt;i++) lp[i]=(i-1)*B+1,rp[i]=i*B; rp[cnt]=n;
  for (int i=1;i<=n;i++) a[pos[i]][++*a[pos[i]]]=i;
  for (int i=1;i<=cnt;i++) sort(a[i]+1,a[i]+*a[i]+1,cmp);
}

int tmp[N]; int pnt;
int tmp1[N],tmp2[N]; int pnt1,pnt2;

inline void Merge(int *a,int na,int *b,int nb,int *tmp){
  pnt=0;
  int p=1,q=1;
  while (p<=na || q<=nb)
    if (p==na+1) tmp[++pnt]=b[q++];
    else if (q==nb+1) tmp[++pnt]=a[p++];
    else if (_a[a[p]]<_a[b[q]]) tmp[++pnt]=a[p++];
    else tmp[++pnt]=b[q++];
}

inline void Add(int l,int r,int w){
  int lb=pos[l],rb=pos[r];
  if (lb==rb){
    for (int i=l;i<=r;i++) _a[i]+=w;
    pnt1=0; pnt2=0;
    for (int i=1;i<=*a[lb];i++)
      if (l<=a[lb][i] && a[lb][i]<=r)
    tmp1[++pnt1]=a[lb][i];
      else
    tmp2[++pnt2]=a[lb][i];
    Merge(tmp1,pnt1,tmp2,pnt2,a[lb]);
    C0+=pnt<<1;
    return;
  }
  for (int i=lb+1;i<rb;i++) tag[i]+=w;
  C0+=rb-lb;

  for (int i=l;i<=rp[lb];i++) _a[i]+=w;

  pnt1=0,pnt2=0;
  for (int i=1;i<=*a[lb];i++)
    if (a[lb][i]>=l) tmp1[++pnt1]=a[lb][i];
    else tmp2[++pnt2]=a[lb][i];
  Merge(tmp1,pnt1,tmp2,pnt2,a[lb]);
  C0+=pnt<<1;

  for (int i=lp[rb];i<=r;i++) _a[i]+=w;

  pnt1=0,pnt2=0;
  for (int i=1;i<=*a[rb];i++)
    if (a[rb][i]<=r) tmp1[++pnt1]=a[rb][i];
    else tmp2[++pnt2]=a[rb][i];
  Merge(tmp1,pnt1,tmp2,pnt2,a[rb]);
  C0+=pnt<<1;
}

inline int Bin(int *a,int n,int w){
  _a[0]=w; C1+=10;
  return upper_bound(a+1,a+n+1,0,cmp)-a-1;
  //  int L=0,R=n,MID;
  //  if (_a[a[n]]<=w) return n;
  //  while (L+1<R){
  //    C1++;
  //    if (_a[a[MID=(L+R)>>1]]>w)
  //      R=MID;
  //    else
  //      L=MID;
  //  }
  //  return L;
}

inline int Query(int l,int r,int K){
  if (K>r-l+1) return -1;
  int lb=pos[l],rb=pos[r];
  if (lb==rb){
    pnt=0;
    for (int i=1;i<=*a[lb];i++)
      if (l<=a[lb][i] && a[lb][i]<=r)
    tmp[++pnt]=a[lb][i];
    C1+=*a[lb];
    return _a[tmp[K]]+tag[lb];
  }
  pnt1=0,pnt2=0;
  for (int i=1;i<=*a[lb];i++)
    if (a[lb][i]>=l) tmp1[++pnt1]=a[lb][i],_a[a[lb][i]]+=tag[lb];
  for (int i=1;i<=*a[rb];i++)
    if (a[rb][i]<=r) tmp2[++pnt2]=a[rb][i],_a[a[rb][i]]+=tag[rb];

  Merge(tmp1,pnt1,tmp2,pnt2,tmp);
  C1+=pnt<<1;

  int L=-1,R=(n+Q)*len;
  while (L+1<R){
    int MID=(L+R)>>1,tot=0;
    if (L<100000 && R>100000) MID=100000;
    for (int i=lb+1;i<rb;i++)
      tot+=Bin(a[i],*a[i],MID-tag[i]);
    //    for (int i=l;i<=rp[lb];i++) tot+=_a[i]+tag[lb]<=MID;
    //    for (int i=lp[rb];i<=r;i++) tot+=_a[i]+tag[rb]<=MID;
    tot+=Bin(tmp,pnt,MID);
    if (tot<K)
      L=MID;
    else
      R=MID;
  }
  for (int i=l;i<=rp[lb];i++) _a[i]-=tag[lb];
  for (int i=lp[rb];i<=r;i++) _a[i]-=tag[rb];
  return R;
}

int main(){
  freopen("j.in","r",stdin);
  freopen("j1.out","w",stdout);
  int x,y,order;
  read(n); read(Q); read(len);
  for (int i=2;i<=n;i++)
    read(x),read(y),add(x,i,y,++inum);
  dfs(1);
  for (int i=1;i<=n;i++) _a[pre[i]]=dis[i];
  Build();
  int c0=0,c1=0;
  for (int t=1;t<=Q;t++){
    read(order); read(x); read(y);
    if (order==1){
      c0++;
      int ans=Query(pre[x],pre[x]+size[x]-1,y);
      printf("%d\n",ans);
    }
    else
      c1++,Add(pre[x],pre[x]+size[x]-1,y);
  }
//  fprintf(stderr,"%d %d %d %d %d %d\n",B,cnt,c0,c1,C0,C1);
  return 0;
}