First of all English is not my native language, so if I make any grammatical mistakes sorry about that, and also this code migh be inefficient but I am newbie to MPI please don’t hit me hard.

This code working without splitting channels however, for now my aim is to scatter first two rows to the processes 0 and 1, and last two rows to the processes 2 and 3. Once I manage this I’ll do reverse logic to the column wise.

If needed I can put whole code in here, please let me know.

I have helper functions for the 2D array:

template <typename T> T **alloc2D(int n, int m) {
  T *data = new T[n * m];
  T **array = new T *[n];
  for (int i = 0; i < n; i++)
    array[i] = &(data[i * m]);

  return array;
}

template <typename T> void free2D(T **array) {
  if (array != nullptr) {
    delete[] array[0];
    delete[] array;
  }
}

File reading:

template <typename T>
void readMatrixFromFile(std::string fileName, T **matrix, int numRows,
                        int numCols) {
  std::ifstream file(fileName);
  if (!file.is_open()) {
    std::cerr << "Failed to open file: " << fileName << std::endl;
    return;
  } else {
    std::string line;
    int row = 0;
    while (std::getline(file, line) && row < numRows) {
      std::istringstream iss(line);
      int col = 0;
      int value;
      while (iss >> value && col < numCols) {
        matrix[row][col] = value;
        col++;
      }
      row++;
    }
  }
}

Two channel creation:

MPI_Comm rowGroupCom;
int rowGroup = rank % 2;
MPI_Comm_split(MPI_COMM_WORLD, rowGroup, rank, &rowGroupCom);

Matrix read and creating ptrs:

  int **matrixA = nullptr;
  int **localA = nullptr;
  int *sendRowsPtr = nullptr;
  int numRowsPerProc = numRows / (size / 2);

  if (rowGroup == 0) {
    matrixA = alloc2D<int>(numRows, numCols);
    readMatrixFromFile("matrixA.txt", matrixA, numRows, numCols);

    if (rank == 0) {
      sendRowsPtr = &(matrixA[0][0]);
    } else {
      sendRowsPtr = &(matrixA[numRowsPerProc][0]);
    }
  } else {
    sendRowsPtr = NULL;
  }

and with the help of this post scattering rows:

  MPI_Datatype rootRow, rootRowType, finalRow, finalRowType;

  if (rowGroup == 0) {
    MPI_Type_contiguous(numCols, MPI_INT, &rootRow);
    MPI_Type_commit(&rootRow);
    MPI_Type_create_resized(rootRow, 0, numCols * sizeof(int), &rootRowType);
    MPI_Type_commit(&rootRowType);
  }

  MPI_Type_contiguous(numCols, MPI_INT, &finalRow);
  MPI_Type_commit(&finalRow);
  MPI_Type_create_resized(finalRow, 0, numCols * sizeof(int), &finalRowType);
  MPI_Type_commit(&finalRowType);

  MPI_Scatter(sendRowsPtr, numRowsPerProc, rootRowType, &(localA[0][0]),
              numRowsPerProc, finalRowType, 0, rowGroupCom);

Freeing some stuffs:

  MPI_Barrier(MPI_COMM_WORLD);
  if (rank == 0) {
    MPI_Type_free(&rootRow);
    MPI_Type_free(&rootRowType);
  }

  MPI_Type_free(&finalRow);
  MPI_Type_free(&finalRowType);

  if (rank == 0) {
    free2D(matrixA);
  }
  free2D(localA);

However when I compile and run it with valgrid I’m getting this:

[endeavourOSLap:108999:0:108999] Caught signal 11 (Segmentation fault: address not mapped to object at address 0xffffffffffffff80)

Rank 0 received row matrix: 
0 1 2 3 
4 5 6 7 

Rank 2 received row matrix: 
8 9 10 11 
12 13 14 15 
==== backtrace (tid: 108999) ====
 0 0x000000000004b5ce ucs_event_set_fd_get()  ???:0
 1 0x000000000004b7aa ucs_event_set_fd_get()  ???:0
 2 0x000000000003cae0 __sigaction()  ???:0
 3 0x0000000000133fe3 ompi_coll_tuned_scatter_intra_dec_fixed()  ???:0
 4 0x00000000000d5729 MPI_Scatter()  ???:0
 5 0x0000000000002702 main()  /home/bestsithineu/Documents/cse574/final/trial.cpp:118
 6 0x0000000000025c88 __libc_init_first()  ???:0
 7 0x0000000000025d4c __libc_start_main()  ???:0
 8 0x00000000000022f5 _start()  ???:0
=================================
--------------------------------------------------------------------------
prterun noticed that process rank 1 with PID 108999 on node endeavourOSLap exited on
signal 11 (Segmentation fault).
--------------------------------------------------------------------------

Desired output:

Rank 0 received row matrix: 
0 1 2 3 
4 5 6 7 
Rank 1 received row matrix: 
0 1 2 3 
4 5 6 7 
Rank 2 received row matrix: 
8 9 10 11 
12 13 14 15
Rank 3 received row matrix: 
8 9 10 11 
12 13 14 15