#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>

#define n 16

int main(int argc, char *argv[])
{
  MPI_Init(&argc, &argv);
  MPI_Comm comm = MPI_COMM_WORLD;
  int p,r;
  MPI_Comm_size(comm, &p);
  MPI_Comm_rank(comm, &r);
  int *arr;
  arr = NULL;
  if (r == 0){
    arr = (int *) malloc(n * n * sizeof(int));
    for (int i = 0; i < n * n; i++) arr[i] = i;
    for (int i = 0; i < n; i++){
      printf("n");
      for (int j = 0; j < n; j++)
        printf("%4d", arr[i * n + j]);
    }
  }
  printf("n");
  int ps = sqrt(p);
  int ns = n / ps;

  if (r == 0) {
    printf("ps: %d ns: %dn", ps, ns);
  }
  /* create datatype */
  MPI_Datatype block;
  MPI_Type_vector(ns, ns, n, MPI_INT, &block);
  int blocks[ps];
  MPI_Aint displs[ps];
  for (int i = 0; i < ps; i++) {
    blocks[i] = 1;
    displs[i] = i  * sizeof(int);
  }
  MPI_Datatype types[ps];
  //for (int i = 0; i < ps - 1; i++) types[i] = block;
  //types[ps - 1] = MPI_UB;
  types[0] = block;
  for (int i = 1; i < ps; i++) types[i] = MPI_UB;
  //types[0] = block;
  //types[1] = MPI_UB;
  if (r == 0) {
    printf("displs:n");
    for(int i = 0; i < ps; i++) printf("%3ld", displs[i]);
    printf("n");
  }

  MPI_Datatype row;
  MPI_Type_struct(ps, blocks, displs, types, &row);
  MPI_Type_commit(&row);

  /* prepare scatter */
  int sdispl[p]; int sendcounts[p];
  for (int i = 0; i < p; i++) {
    sdispl[i] = (i % ps) + (i / ps) * (ns * ps);
    sendcounts[i] = 1;
  }
  if (r == 0) {
    printf("sdispl: n");
    for (int i = 0; i < 4; i++) printf("%3d", sdispl[i]);
    printf("n");
  }

  int rcv[ns * ns];
  MPI_Scatterv(arr, sendcounts, sdispl, row, rcv, ns * ns, MPI_INT, 0, comm);

  int result = 1;
  if (r == result) {
    printf("result for %d:n", result);
    for (int i = 0; i < ns * ns; i++) {
      printf("%4d", rcv[i]);
      if ((i+1) % ns == 0) printf("n");
    }
  }

  if (arr != NULL) free(arr);
  MPI_Finalize();
  return 0;
}