英文:
Why is sort with std::execution::par is taking longer than regular sort?
问题
为什么相同的向量v1（通常情况下）和v2（使用std::execution:parallel）对于v1花费的时间较短，而对于v2花费的时间较长？
我期望并行排序完成得更快。
结果:
626600 = 串行计数
1048100 = 并行计数
英文:
Why does the identical vectors v1 (normally) and v2 (with std::execution:parallel) takes less time for v1 and more time for v2?
I was expecting the parallel sort to complete faster.
#include <iostream>
#include <vector>
#include <chrono>
#include <algorithm>
#include <execution>
using namespace std;
int main() {
// Generate identical vectors v1 and v2 of length N
// and fill them with v[i] = i/200;
int N = 10000;
vector<int> v1(N);
for(int i=0; i<N; i++) v1[i] = i/200;
auto v2 = v1;
// Calculate the time taken to sort in serial way.
auto start = std::chrono::system_clock::now();
sort(v1.begin(), v1.end());
auto end = std::chrono::system_clock::now();
auto diff = end - start;
cout << diff.count() << " = serial count" << endl;
// Calculate the time taken to sort in parallel way using
// std::execution::par
start = std::chrono::system_clock::now();
sort(std::execution::par, v2.begin(), v2.end());
end = std::chrono::system_clock::now();
diff = end - start;
cout << diff.count() << " = parallel parallel" << endl;
return 0;
}
Result:
626600 = serial count
1048100 = parallel count
答案1
得分: 3
int N = 10000; 是一个对于使用 std::execution::par 来说元素数量太小的情况。对于这么少的元素，运行和等待线程完成的开销可能会超过串行运行的时间。无论如何，我在 onlinegdb.com 上得到了不同的结果（使用 -O2 编译选项）：
134692 = 串行计数
128443 = 并行计数
英文:
int N = 10000; is too small amount of elements for involving std::execution::par. Overheads for running and waiting for finishing threads can exceed serial run time with such a small amount of elements. In any case I get other results on onlinegdb.com (with -O2):
134692 = serial count
128443 = parallel count
通过集体智慧和协作来改善编程学习和解决问题的方式。致力于成为全球开发者共同参与的知识库，让每个人都能够通过互相帮助和分享经验来进步。
评论