在C ++中对向量进行并行操作的最快方法？

Question

I am trying to parallelize operations over a large vector of objects in C++. I have written parallel programs in Java before, but I have just started using C++.

The current code uses an iterator over the vector. What would be the fastest way to parallelize this? My current thoughts are...

1. Using the .size() function and using a forloop through the vector. However, I am worried about the runtime of the .size() function, is it O(N) or O(1)? Also would forloops be slower than using an iterator?

2. Somehow splitting the vector, and create iterators for the new vectors in parallel? If so, what would be a good method of splitting the vector with a fast runtime?

Or is there some faster way to do this?

Answer 1

However, I am worried about the runtime of the .size() function, is it O(N) or O(1)?

vector<...>::size() is O(1).

Also would forloops be slower than using an iterator?

In most cases, I doubt it. In some cases, an algorithm which takes iterators may be optimized based on the value type of the iterator. Benchmark it.

Somehow splitting the vector, and create iterators for the new vectors in parallel? If so, what would be a good method of splitting the vector with a fast runtime?

Vector iterators are random access. It is a very cheap operation to just find the distance from begin to end (O(1)), and split it in half.

auto begin = v.begin();
auto end = v.end();
auto mid = begin + (end - begin)/2;
algorithm(begin, mid);
algorithm(mid, end);

Answer 2

如果向量没有被迭代器修改，我会将向量划分为N个部分，并为每个线程分配其向量，然后，如果需要，一个线程会汇总所有其他线程的结果。

Answer 3

look into boost threads. I find them very intuitive

Answer 4

我相信这个答案是有意义的： https : //stackoverflow.com/a/2547725/1524700根据您的平台，我还将英特尔线程构建模块添加到列表中。