BOJ 2357 최솟값과 최댓값

simd_practice_2.cpp

#include <bits/stdc++.h>
#include <immintrin.h>
#pragma GCC target("avx,avx2,fma")
using namespace std;

typedef long long ll;

int n, q;
alignas(256) int a[101010];
const int inf = 1e9+7;

void query(int s, int e){
    int mx = -inf, mn = inf, i;
    __m256i avxMax = _mm256_set1_epi32(-inf);
    __m256i avxMin = _mm256_set1_epi32(inf);
    for(i=s; (i&7) && i<=e; i++) mx = max(mx, a[i]), mn = min(mn, a[i]);
    for(; i+8<e; i+=8){
        __m256i now = _mm256_load_si256((__m256i const *)(a+i));
        avxMax = _mm256_max_epi32(avxMax, now);
        avxMin = _mm256_min_epi32(avxMin, now);
    }
    for(; i<=e; i++) mx = max(mx, a[i]), mn = min(mn, a[i]);
    int t1[8], t2[8];
    _mm256_storeu_si256((__m256i *)t1, avxMax);
    _mm256_storeu_si256((__m256i *)t2, avxMin);
    for(i=0; i<8; i++) mx = max(mx, t1[i]), mn = min(mn, t2[i]);
    cout << mn << " " << mx << "\n";
}

int main(){
    ios_base::sync_with_stdio(false); cin.tie(nullptr);
    //cin >> n >> q;
    n = q = 100000;
    for(int i=0; i<n; i++) a[i] = rand();
    for(int i=0; i<q; i++){
        int a, b; //cin >> a >> b;
        a = 1 + (i & 31);
        b = 100000 - (i & 31);
        query(a-1, b-1);
    }
}