Duplicate Messages on Matrix?

[KIwabuchi]

When I used the Matrix cluster, I got duplicate ygm messages when sending messages with std::vector data.
I don't see the same error on Dane. Do you have any ideas on what might be causing this, or how to fix it?

I used 2 nodes and 8 ranks.
I don't get the error if I use only one node or a small vector size (such as 4).
As for the compiler, I just loaded GCC 13 (ml gcc/13.3.1-magic).
I tried the master and dev-v0.9 branches.

Here is an example code:

#include <iostream>
#include <unordered_set>
#include <vector>
#include <ygm/comm.hpp>

int main(int argc, char** argv) {
  ygm::comm comm(&argc, &argv);

  static std::unordered_set<uint64_t> table;
  comm.cf_barrier();

  constexpr int vec_size = 128;
  std::vector<float> data(vec_size, 0.0f);

  constexpr uint64_t chunk_size = 1 << 20;
  for (uint64_t i = 0; i < chunk_size; ++i) {
    const uint64_t id = i + comm.rank() * chunk_size;

    auto receiver = [](auto, const uint64_t id,
                       const std::vector<float>& data) {
      if (table.contains(id)) {
        std::cerr << "Duplicate ID " << id << std::endl;
        MPI_Abort(MPI_COMM_WORLD, EXIT_FAILURE);
      }
      table.insert(id);
    };
    comm.async(id % comm.size(), receiver, id, data);
  }
  comm.barrier();

  std::cout << "All done " << comm.rank() << std::endl;
  comm.cf_barrier();

  return 0;
}

Thanks!
Keita

[steiltre]

I have not seen this before. I'll take a look into it.