capsule-ft-mpi.h

MPI synchronization of the Lagrangian mesh

MPI synchronization of the Lagrangian mesh

In the current implementation, each processor has its own copy of the Lagrangian mesh. In order to update the position of all the local copies of the Lagrangian mesh, the processors need to communicate to each other the contributions of their flow field to the velocities of each Lagrangian nodes.

void reduce_lagVel(lagMesh* mesh) {
  int li = dimension; // length of one item
  double* send_data = (double*)malloc(mesh->nln*li*sizeof(double));
  double* recv_data = (double*)malloc(mpi_npe*mesh->nln*li*sizeof(double));
  for(int i=0; i<mesh->nln; i++) {
    send_data[li*i] = mesh->nodes[i].lagVel.x;
    send_data[li*i+1] = mesh->nodes[i].lagVel.y;
    #if dimension > 2
    send_data[li*i+2] = mesh->nodes[i].lagVel.z;
    #endif
  }
  MPI_Allgather(send_data, mesh->nln*li, MPI_DOUBLE, recv_data,
    mesh->nln*li, MPI_DOUBLE, MPI_COMM_WORLD);

  for(int i=0; i<mesh->nln; i++) {
    mesh->nodes[i].lagVel.x = 0.;
    mesh->nodes[i].lagVel.y = 0.;
    #if dimension > 2
    mesh->nodes[i].lagVel.z = 0.;
    #endif
  }

  for(int k=0; k<mpi_npe; k++) {
    for(int i=0; i<mesh->nln; i++) {
      mesh->nodes[i].lagVel.x += recv_data[k*li*mesh->nln+li*i];
      mesh->nodes[i].lagVel.y += recv_data[k*li*mesh->nln+li*i+1];
      #if dimension > 2
      mesh->nodes[i].lagVel.z += recv_data[k*li*mesh->nln+li*i+2];
      #endif
    }
  }
  free(send_data);
  free(recv_data);
}