Seg fault: c++17 std parallel for_each and DynamicMatrix

I’ve been testing a use-case, example code goes as follows,

#include <blaze/Blaze.h>
#include <execution>
#include <iostream>
#include <vector>

int main(int argc, char** argv) {
  // determine how many workers run in parallel
  std::vector<size_t> workers(std::stoi(argv[1]));

  std::for_each(std::execution::par, workers.begin(), workers.end(), [&](size_t x) {
    blaze::DynamicMatrix<float> mat;
    mat.resize(10000, 50);
    mat = 1.0f;

    for (size_t i = 0; i < blaze::rows(mat); ++i) {
      // This line seems to be the place where segfault happens
      blaze::row(mat, i) = blaze::row(mat, 0);
    }
    std::cout << blaze::sum(mat) << std::endl;
  });
}

‌

Env:

Mac OSX Catalina, 10.15.7.
C++ compiler is GNU G++-10.

Compiled like this,

g++ -std=c++17 -O3 -DNDEBUG a.cpp -ltbb -I /usr/local/include -IX/ -L /usr/local/lib -fopenmp

If run with 1 worker, all is good, no segfault.

But if run with >1 workers, segfault happens, but not constantly.

>>> ./a.out 2
500000
terminate called after throwing an instance of 'std::runtime_error'
zsh: segmentation fault  ./a.out 2

‌

line14 seems to be the cause, since if I comment out it, the code won’t seg fault.

The problem is I don’t understand why that line would be an issue?

Comments (3)