admin管理员组

文章数量:1352141

In my projects, there are several std::vector<std::vector<int>> dynamic arrays. The first dimension is about 10 million, the second dimension is about 10-100. Both of the dimensions are dynamic, i.e. frequently push_back/resize/clear, which seriously affects the performance of the parallel logic, such as OpenMP and TBB. In some cases, the CPU utilization is lower to 20% with the full kernel added.

How to solve this problem?

std::vector<std::vector<int>> A;
A.clear();
A.resize(12000000);
for(int32_t i=0; i<A.size(); ++i)
{
    A[i].resize(32);
    for(int32_t j=0; j<32; j++)
    {
        A[i].push_back(i * j);
    }
}

I have tried to use std::pmr::vector<std::pmr::vector<int>> to replace the original nested vector, however the performance of the parallel logic is affected as well.

Moreover, I have tried using std::vector<std::vector<int, boost::pool_allocator<int>>>, however the resize/clear of such vector is too inefficient, and the program seems to be hung.

本文标签： cNested stdvectorltstdvectorgt brings too many memory fragmentation and pollutes the cacheStack Overflow

版权声明：本文标题：c++ - Nested std::vector&lt;std::vector&gt; brings too many memory fragmentation and pollutes the cache - Stack Overflow 内容由网友自发贡献，该文观点仅代表作者本人， 转载请联系作者并注明出处：http://www.betaflare.com/web/1743910563a2560321.html， 本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌抄袭侵权/违法违规的内容，一经查实，本站将立刻删除。