#include<bits/stdc++.h>
#include<immintrin.h>
#include<smmintrin.h>
#pragma GCC target("avx2,sse4")
using namespace std;
int N, Q;
int S[100004], T[100004], U[100004];
int x[8], y[8], z[8], b[8];
int main()
{
cin.sync_with_stdio(false);
cout.sync_with_stdio(false);
cin.tie(0); cout.tie(0);
cin >> N >> Q;
for(int i=0; i<N; ++i)
{
cin >> S[i] >> T[i];
U[i] = S[i] + T[i];
}
int rawN = N;
N = (N+7)>>3;
for(int i=0; i<Q; ++i)
{
int X, Y, Z;
cin >> X >> Y >> Z;
if(X==0&&Y==0&&Z==0)
{
cout << rawN << '\n';
continue;
}
for(int j=0; j<8; ++j) x[j]=X,y[j]=Y,z[j]=Z;
__m256i ans = _mm256_load_si256((__m256i*)b);
__m256i xv = _mm256_load_si256((__m256i*)x);
__m256i yv = _mm256_load_si256((__m256i*)y);
__m256i zv = _mm256_load_si256((__m256i*)z);
int ansv = N << 3;
for(int j=0; j<N; ++j)
{
__m256i sv = _mm256_load_si256( ((__m256i*)S)+j);
__m256i tv = _mm256_load_si256( ((__m256i*)T)+j);
__m256i uv = _mm256_load_si256( ((__m256i*)U)+j);
__m256i sx = _mm256_cmpgt_epi32(xv, sv); //xv>sv
__m256i ty = _mm256_cmpgt_epi32(yv, tv); //yv>tv
__m256i uz = _mm256_cmpgt_epi32(zv, uv); //zv>uv
__m256i st = _mm256_or_si256(sx, ty);
__m256i av = _mm256_or_si256(st, uz);
ans = _mm256_add_epi32(ans, av);
}
__m256i ans2 = ans;
int* av = (int*)&ans2;
for(int i=0; i<8; ++i) ansv += av[i];
cout << ansv << '\n';
}
return 0;
}
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
3 ms |
380 KB |
Output is correct |
2 |
Correct |
2 ms |
376 KB |
Output is correct |
3 |
Correct |
2 ms |
376 KB |
Output is correct |
4 |
Correct |
2 ms |
380 KB |
Output is correct |
5 |
Correct |
2 ms |
376 KB |
Output is correct |
6 |
Correct |
2 ms |
376 KB |
Output is correct |
7 |
Correct |
7 ms |
504 KB |
Output is correct |
8 |
Correct |
7 ms |
632 KB |
Output is correct |
9 |
Correct |
7 ms |
632 KB |
Output is correct |
10 |
Correct |
6 ms |
504 KB |
Output is correct |
11 |
Correct |
6 ms |
632 KB |
Output is correct |
12 |
Correct |
5 ms |
376 KB |
Output is correct |
13 |
Correct |
6 ms |
632 KB |
Output is correct |
14 |
Correct |
7 ms |
508 KB |
Output is correct |
15 |
Correct |
7 ms |
504 KB |
Output is correct |
16 |
Correct |
6 ms |
504 KB |
Output is correct |
17 |
Correct |
6 ms |
504 KB |
Output is correct |
18 |
Correct |
5 ms |
376 KB |
Output is correct |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Execution timed out |
3014 ms |
4492 KB |
Time limit exceeded |
2 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Execution timed out |
3014 ms |
4492 KB |
Time limit exceeded |
2 |
Halted |
0 ms |
0 KB |
- |
# |
결과 |
실행 시간 |
메모리 |
Grader output |
1 |
Correct |
3 ms |
380 KB |
Output is correct |
2 |
Correct |
2 ms |
376 KB |
Output is correct |
3 |
Correct |
2 ms |
376 KB |
Output is correct |
4 |
Correct |
2 ms |
380 KB |
Output is correct |
5 |
Correct |
2 ms |
376 KB |
Output is correct |
6 |
Correct |
2 ms |
376 KB |
Output is correct |
7 |
Correct |
7 ms |
504 KB |
Output is correct |
8 |
Correct |
7 ms |
632 KB |
Output is correct |
9 |
Correct |
7 ms |
632 KB |
Output is correct |
10 |
Correct |
6 ms |
504 KB |
Output is correct |
11 |
Correct |
6 ms |
632 KB |
Output is correct |
12 |
Correct |
5 ms |
376 KB |
Output is correct |
13 |
Correct |
6 ms |
632 KB |
Output is correct |
14 |
Correct |
7 ms |
508 KB |
Output is correct |
15 |
Correct |
7 ms |
504 KB |
Output is correct |
16 |
Correct |
6 ms |
504 KB |
Output is correct |
17 |
Correct |
6 ms |
504 KB |
Output is correct |
18 |
Correct |
5 ms |
376 KB |
Output is correct |
19 |
Execution timed out |
3014 ms |
4492 KB |
Time limit exceeded |
20 |
Halted |
0 ms |
0 KB |
- |