/*
 * Copyright(C) 1999-2020 National Technology & Engineering Solutions
 * of Sandia, LLC (NTESS).  Under the terms of Contract DE-NA0003525 with
 * NTESS, the U.S. Government retains certain rights in this software.
 *
 * See packages/seacas/LICENSE for details
 */

#include "structs.h"

/* Combine goals of collections of processors for next division. */
void merge_goals(double *         goal,         /* desired set sizes */
                 double *         merged_goal,  /* sizes of sets at this partition level */
                 struct set_info *set_info,     /* information about all sets */
                 int *            subsets,      /* set numbers of processors to merge */
                 int              nsets,        /* number of sets created by this division */
                 int              ndims_tot,    /* total number of dimensions in the hypercube */
                 int              cube_or_mesh, /* 0=> hypercube, d=> d-dimensional mesh */
                 int              mesh_dims[3], /* shape of mesh */
                 double           vwgt_sum      /* actual sum of vertex weights */
)
{
  struct set_info *set;        /* set of processors still clumped together */
  double           total_goal; /* total of desired goals */
  int              index;      /* x, y or z location of a processor */
  int              i, x, y, z; /* loop counters */

  total_goal = 0;
  for (i = 0; i < nsets; i++) {
    set            = &(set_info[subsets[i]]);
    merged_goal[i] = 0;

    if (cube_or_mesh > 0) { /* Mesh architecture. */
      for (x = set->low[0]; x < set->low[0] + set->span[0]; x++) {
        for (y = set->low[1]; y < set->low[1] + set->span[1]; y++) {
          for (z = set->low[2]; z < set->low[2] + set->span[2]; z++) {
            index = z * mesh_dims[0] * mesh_dims[1] + y * mesh_dims[0] + x;
            merged_goal[i] += goal[index];
          }
        }
      }
    }

    else if (cube_or_mesh == 0) { /* Hypercube architecture. */
      x = 1 << (ndims_tot - set->ndims);
      y = 1 << ndims_tot;
      for (z = set->setnum; z < y; z += x) {
        merged_goal[i] += goal[z];
      }
    }

    total_goal += merged_goal[i];
  }

  /* Now scale goals to reflect actual weight of vertices available. */
  for (i = 0; i < nsets; i++) {
    merged_goal[i] = (merged_goal[i] / total_goal) * vwgt_sum;
  }
}