send and receive array in MPI

Question

I am new to MPI and I am writing a simple MPI program to get the dot product of matrix and vector, namely, A*b=c. However, my code doesn't work. The source code is listed below.

If I replace the declaration of A, b, c and buffer by

double A[16], b[4], c[4], buffer[8];

and comment those lines related to allocate and free operations, my code works and the result is correct. In this case, I was wondering that the trouble should be related to pointers, but I have no idea to shoot the trouble.

One more thing, in my code, buffer has only 4 elements, but the buffer size must greater than 8, or it doesn't work.

#include<mpi.h>
#include<iostream>
#include<stdlib.h>

using namespace std;

int nx = 4, ny = 4, nxny;
int ix, iy;
double *A = nullptr, *b = nullptr, *c = nullptr, *buffer = nullptr;
double ans;

// info MPI
int myGlobalID, root = 0, numProc;
int numSent;
MPI_Status status;

// functions
void get_ixiy(int);

int main(){

  MPI_Init(NULL, NULL);
  MPI_Comm_size(MPI_COMM_WORLD, &numProc);
  MPI_Comm_rank(MPI_COMM_WORLD, &myGlobalID);

  nxny = nx * ny;

  A = new double(nxny);
  b = new double(ny);
  c = new double(nx);
  buffer = new double(ny);

  if(myGlobalID == root){
    // init A, b
    for(int k = 0; k < nxny; ++k){
      get_ixiy(k);
      b[iy] = 1;
      A[k] = k;
    }
    numSent = 0;

    // send b to each worker processor
    MPI_Bcast(&b, ny, MPI_DOUBLE, root, MPI_COMM_WORLD);

    // send a row of A to each worker processor, tag with row number
    for(ix = 0; ix < min(numProc - 1, nx); ++ix){
      for(iy = 0; iy < ny; ++iy){
        buffer[iy] = A[iy + ix * ny];
      }
      MPI_Send(&buffer, ny, MPI_DOUBLE, ix+1, ix+1, MPI_COMM_WORLD);
      numSent += 1;
    }

    for(ix = 0; ix < nx; ++ix){
      MPI_Recv(&ans, 1, MPI_DOUBLE, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, &status);
      int sender = status.MPI_SOURCE;
      int ansType = status.MPI_TAG;
      c[ansType] = ans;

      // send another row to worker process
      if(numSent < nx){
        for(iy = 0; iy < ny; ++iy){
          buffer[iy] = A[iy + numSent * ny];
        }
        MPI_Send(&buffer, ny, MPI_DOUBLE, sender, numSent+1, 
        MPI_COMM_WORLD);
        numSent += 1;
      }
      else
        MPI_Send(MPI_BOTTOM, 0, MPI_DOUBLE, sender, 0, MPI_COMM_WORLD);
    }

    for(ix = 0; ix < nx; ++ix){
      std::cout << c[ix] << " ";
    }
    std::cout << std::endl;

    delete [] A;
    delete [] b;
    delete [] c;
    delete [] buffer;
  }
  else{
    MPI_Bcast(&b, ny, MPI_DOUBLE, root, MPI_COMM_WORLD);
      if(myGlobalID <= nx){
        while(1){
          MPI_Recv(&buffer, ny, MPI_DOUBLE, root, MPI_ANY_TAG, MPI_COMM_WORLD, &status);
          if(status.MPI_TAG == 0) break;
          int row = status.MPI_TAG - 1;
          ans = 0.0;

          for(iy = 0; iy < ny; ++iy) ans += buffer[iy] * b[iy];

          MPI_Send(&ans, 1, MPI_DOUBLE, root, row, MPI_COMM_WORLD);
      }
    }
  }

  MPI_Finalize();
  return 0;
} // main

void get_ixiy(int k){
  ix = k / ny;
  iy = k % ny;
}

The error information is listed below.

=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   PID 7455 RUNNING AT ***
=   EXIT CODE: 11
=   CLEANING UP REMAINING PROCESSES
=   YOU CAN IGNORE THE BELOW CLEANUP MESSAGES

YOUR APPLICATION TERMINATED WITH THE EXIT STRING: Segmentation fault: 
11 (signal 11)
This typically refers to a problem with your application.
Please see the FAQ page for debugging suggestions

There is an array overflow when you b[k] = 1. I recommend you keep your code simple and declare double A[4][4] — Gilles Gouaillardet
– Gilles Gouaillardet, Commented Jun 1, 2018 at 10:36
@GillesGouaillardet Thank you very much for your comment, I have corrected this line, but still problematic. The reason why I want to declare double *A is that I want to deal with dynamic arrays in the future. — Y.Zhou
– Y.Zhou, Commented Jun 1, 2018 at 11:48
If you send/recv/bcast an array, the buffer is the array and not its address. For example, it should be MPI_Bcast(b, ...) and not ...(&b) — Gilles Gouaillardet
– Gilles Gouaillardet, Commented Jun 1, 2018 at 12:31
@GillesGouaillardet Thank you for this comment, I corrected this error and the code works. Thanks a lot! — Y.Zhou
– Y.Zhou, Commented Jun 1, 2018 at 12:40

Ali · Accepted Answer · 2018-06-01 12:36:31Z

1

There are several problems in your code, that you have to fix first.

First, You want to access an element of b[] which do not exist, in this for loop:

for(int k = 0; k < nxny; ++k){
  get_ixiy(k);
  b[k] = 1;     // WARNING: this is an error
  A[k] = k;
}

Second, You are deleting allocated memory only for the root process. This causes a memory leak:

if(myGlobalID == root){
  // ...
  delete [] A;
  delete [] b;
  delete [] c;
  delete [] buffer;
}

You have to to delete the allocated memory for all processes.

Third, You have a useless function void get_ixiy(int); that changes global variables ix, iy. It is useless because, after calling this function, you never use ix, iy until you change them manually. See here:

for(ix = 0; ix < min(numProc - 1, nx); ++ix){
    for(iy = 0; iy < ny; ++iy){
        // ...
    }
}

Fourth, you are using MPI_Send() and MPI_Recv() in a completely wrong way. You are lucky that you don't get more errors.

edited Jun 1, 2018 at 12:36

answered Jun 1, 2018 at 12:08

Ali

1,0731 gold badge13 silver badges33 bronze badges

Sign up to request clarification or add additional context in comments.

1 Comment

Y.Zhou Over a year ago

Thank you for your comments, they are helpful.

Collectives™ on Stack Overflow

send and receive array in MPI

1 Answer 1

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related