根据题意卡常数
#include <bits/stdc++.h> const long long mod = 1e9+7; const double ex = 1e-10; #define inf 0x3f3f3f3f using namespace std; int n,m; unsigned A,B,C; unsigned x,y,z; map <int,unsigned> M; unsigned rng61() { unsigned t; x ^= x << 16; x ^= x >> 5; x ^= x << 1; t = x; x = y; y = z; z = t ^ x ^ y; return z; } unsigned a[11234567]; struct querry { int id; int q; unsigned ans; }b[200]; bool cmpq(querry x,querry y) { return x.q > y.q; } bool cmpid(querry x,querry y) { return x.id < y.id; } int part(int low, int high) { swap(a[high],a[(low+high)/2]); unsigned pivot = a[high]; while(low < high){ while(low < high&&a[low] <= pivot) low++; a[high] = a[low]; while(low < high&&a[high] >= pivot) high--; a[low] = a[high]; } a[high]=pivot; return high; } unsigned quicksort(int l, int r, int k){ int pos = part(l,r); if(pos == k) return a[pos]; else if(pos > k) return quicksort(l,pos-1,k); else return quicksort(pos+1,r,k); } int main() { int cas = 0; while (~scanf("%d%d%u%u%u",&n,&m,&A,&B,&C)) { M.clear(); x = A, y = B, z = C; for (int i = 1; i<=n; i++) { a[i] = rng61(); } for (int i = 1; i<=m; i++) { scanf("%d",&b[i].q); b[i].id = i; } sort(b+1,b+m+1,cmpq); int last = n; for (int i = 1; i <= m;i++) { if (M.find(b[i].q+1)!=M.end()) b[i].ans = M[b[i].q+1]; else b[i].ans = quicksort(1,last,b[i].q+1); M[b[i].q+1] = b[i].ans; last = b[i].q+1; } sort(b+1,b+m+1,cmpid); printf("Case #%d:",++cas); for (int i = 1; i<=m;i++) printf(" %u",b[i].ans) ; puts(""); } return 0; }
从最大的取值到最小的取值依次使用近似线性复杂度的求第 kk 小的方法即可,该方法的思想与快排相似,可以保证前 k - 1k−1 小的元素都放置在第 kk 小元素的前面,这样枚举的时候就可以依次减少每次的枚举量。