Segmentation fault error using MPI in C?

Question

So I need to implement a vector addition function in parallel using MPI in C. Unfortunately, when I run it, it prints a trace of lots of memory locations and then this message:

==================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   PID 2419 RUNNING AT hbaum-pc
=   EXIT CODE: 6
=   CLEANING UP REMAINING PROCESSES
=   YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
===================================================================================
YOUR APPLICATION TERMINATED WITH THE EXIT STRING: Aborted (signal 6)
This typically refers to a problem with your application.
Please see the FAQ page for debugging suggestions

Here is my function code:

double* vector_vector_addition_parallel(double* a, double* b, int length)
{
  int rank, size, error;
  double* result = (double*)malloc(sizeof(double)*length);

  error = MPI_Init(NULL,NULL);
  error = MPI_Comm_size(MPI_COMM_WORLD,&size);
  error = MPI_Comm_rank(MPI_COMM_WORLD,&rank);

  int sublist_length = (int)(length/size);

  double* sub_a = (double*)malloc(sizeof(double) * sublist_length);
  double* sub_b = (double*)malloc(sizeof(double) * sublist_length);


  error = MPI_Scatter(a, sublist_length, MPI_DOUBLE, sub_a, sublist_length, MPI_DOUBLE, 0, MPI_COMM_WORLD);
  error = MPI_Scatter(b, sublist_length, MPI_DOUBLE, sub_b, sublist_length, MPI_DOUBLE, 0, MPI_COMM_WORLD);

  double* buffer = (double*)malloc(sizeof(double)*sublist_length);
  for(int i = 0; i < sublist_length; i++)
  {
    buffer[i] = sub_a[i] + sub_b[i];
  }

  error = MPI_Gather(buffer,sublist_length,MPI_DOUBLE,result,length,MPI_DOUBLE,0,MPI_COMM_WORLD);

  error = MPI_Finalize();

  return result;
}

And here is the code that calls it:

int main(int argc, char** argv)
{
  double a[8] = {1.0,3.0,5.0,7.0,9.0,11.0,13.0,15.0};
  double b[8] = {2.0,4.0,6.0,8.0,10.0,12.0,14.0,16.0};
  double* vec = vector_vector_addition_parallel(a,b,8);
  return 0;
}

I compile it with

mpicc <source_file>

and run it with

mpiexec -n 4 <path_to_executable>

I also tried using gdb to debug the code but according to gdb, there's no problem and it's absolutely fine. When I use printf to output the vector, it even prints the correct vector when I run the program via gdb.

I think there's something wrong with how I've used MPI_Gather as when I comment it out, the code will run without a segmentation fault although obviously not the correct answer as I need to use MPI_Gather in order to get a result.

How can I improve my code in order to not get a segmentation fault?

Answer 1

Indeed, your MPI_Gather is the issue. The receive count parameter should be the number of elements received from any single process. Thus, you should pass sublist_length instead of length for the recvcount, ie:

MPI_Gather(buffer,sublist_length,MPI_DOUBLE,result,sublist_length,MPI_DOUBLE,0,MPI_COMM_WORLD);