nesting thread pool not work

[wyf]

Describe the new feature

I want to parallel a series code ,it has two for loop as following, where rows is a vector<vector<size_t>>.

        // tmr.start();
        for (i = 0; i < num_rows; i++)
        {
            double h = 0.0;
            for (ci = rows[i].begin(); ci != rows[i].end(); ci++)
            {
                double h_v = (num_outgoing[*ci])
                                 ? 1.0 / static_cast<double>(num_outgoing[*ci])
                                 : 0.0;
                if (num_iterations == 0 && trace)
                {
                    cout << "h[" << i << "," << *ci << "]=" << h_v << endl;
                }
                h += h_v * old_pr[*ci];
            }
            h *= alpha;
            pr[i] = h + one_Av + one_Iv;
            diff += fabs(pr[i] - old_pr[i]);
        }

Code example

I create two independent BS::thread_pool_light variable, here is my parallel code:

        auto loopDiff = [&](const size_t start, const size_t end)
        {
            for (size_t k = start; k < end; k++)
            {
                for (size_t j = 0; j < vectors.at(k).size(); j++)
                {
                    size_t i = vector_size * k + j;
                    if (i >= num_rows)
                        break;
                    double h = 0.0;
                    size_t nb_cnt = rows.at(i).size();
                    size_t vector_size_nb = nb_cnt / nproc + 1;
                    vector<double> vectorH(nb_cnt);
                    auto loopH = [&](const size_t startH, const size_t endH)
                    {
                        for (size_t iH = startH; iH < endH; iH++)
                        {
                            for (size_t jH = 0; jH < vector_size_nb; jH++)
                            {
                                size_t idxH = vector_size_nb * iH + jH;
                                if (idxH >= nb_cnt)
                                    break;
                                auto outEdge = rows.at(i).at(idxH);
                                double h_v = (num_outgoing.at(outEdge))
                                                 ? 1.0 / static_cast<double>(num_outgoing.at(outEdge))
                                                 : 0.0;
                                if (num_iterations == 0 && trace)
                                    cout << "h[" << i << "," << outEdge << "]=" << h_v << endl;
                                vectorH.at(iH) += h_v * old_pr.at(outEdge);
                            }
                        }
                    };
                    pool_neighbor.push_loop(nproc, loopH);
                    pool_neighbor.wait_for_tasks();
                    for (auto hh : vectorH)
                        h += hh;
                    h *= alpha;
                    pr[i] = h + one_Av + one_Iv;
                    vectorDiff.at(i) = fabs(pr[i] - old_pr[i]);
                }
            }
        };
        pool.push_loop(nproc, loopDiff);
        pool.wait_for_tasks();
        for (auto d : vectorDiff)
            diff += d;

Additional information

It does not have parallel acceleration, could you please give me a little advice, thank you .

[Barak Shoshany]

It looks like you are parallelizing a loop where each block in the loop then itself parallelizes and waits for another loop using a separate thread pool. Instead you should only have one thread pool, and parallelize only one loop at a time. If you have any further questions, please post on StackOverflow or StackExchange Code Review. Good luck!