Пример #1
0
int
Zoltan_Matrix2d_adjproc (ZZ* zz, const Zoltan_matrix_2d * const mat, int **adjproc)
{
  static char *yo = "Zoltan_Matrix2d_adjproc";
  int ierr = ZOLTAN_OK;
  int i;
  ZOLTAN_TRACE_ENTER(zz, yo);

  *adjproc = (int*) ZOLTAN_MALLOC(mat->mtx.nPins*sizeof(int));
  if (mat->mtx.nPins && (*adjproc == NULL))
    MEMORY_ERROR;

  for (i = 0 ; i < mat->mtx.nPins ; ++i ) {
    (*adjproc)[i] = EDGE_TO_PROC_Y(mat, mat->mtx.pinGNO[i]);
  }

 End:
  ZOLTAN_TRACE_EXIT(zz, yo);

KDDKDDKDD(zz->Proc, "Zoltan_Matrix2d_adjproc done");
  return (ierr);
}
Пример #2
0
int Zoltan_Color_Test(
    ZZ *zz,                   /* Zoltan structure */
    int *num_gid_entries,     /* # of entries for a global id */
    int *num_lid_entries,     /* # of entries for a local id */
    int num_obj,              /* Input: number of objects */
    ZOLTAN_ID_PTR global_ids, /* Input: global ids of the vertices */
			      /* The application must allocate enough space */
    ZOLTAN_ID_PTR local_ids,  /* Input: local ids of the vertices */
			      /* The application must allocate enough space */
    int *color_exp            /* Input: Colors assigned to local vertices */
)
{
  static char *yo = "color_test_fn";
  int nvtx = num_obj;               /* number of vertices */
  int i, j;
  int ierr = ZOLTAN_OK;
  int ferr = ZOLTAN_OK;             /* final error signal */
  char coloring_problem;   /* Input: which coloring to perform;
			   currently only supports D1, D2 coloring and variants */
  char coloring_problemStr[MAX_PARAM_STRING_LEN]; /* string version coloring problem name */
  int ss=100;
  char comm_pattern='S', coloring_order='I', coloring_method='F';
  int comm[2],gcomm[2];
  int *color=NULL;

  int *adjproc=NULL, *xadj=NULL;
  ZOLTAN_GNO_TYPE gvtx;                         /* number of global vertices */
  ZOLTAN_GNO_TYPE *vtxdist=NULL, *adjncy=NULL;
  ZG graph;
  ZOLTAN_GNO_TYPE *requested_GNOs = NULL;  /* Return GNOs of 
                                              the requested GIDs.  */
  int *loc_partialD2 = NULL;    /* local binary array showing which vertices to be colored */
  int *partialD2 = NULL;        /* global binary array showing which vertices to be colored */
  struct Zoltan_DD_Struct *dd_color;  /* DDirectory for colors */
  ZOLTAN_GNO_TYPE *local_GNOs = NULL;
  ZOLTAN_GNO_TYPE *global_GNOs = NULL;
  
  memset (&graph, 0, sizeof(ZG));

  /* PARAMETER SETTINGS */
  Zoltan_Bind_Param(Color_params, "COLORING_PROBLEM",   (void *) &coloring_problemStr);
  Zoltan_Bind_Param(Color_params, "SUPERSTEP_SIZE",     (void *) &ss);
  Zoltan_Bind_Param(Color_params, "COMM_PATTERN",       (void *) &comm_pattern);
  Zoltan_Bind_Param(Color_params, "VERTEX_VISIT_ORDER", (void *) &coloring_order);
  Zoltan_Bind_Param(Color_params, "COLORING_METHOD",    (void *) &coloring_method);

  strncpy(coloring_problemStr, "distance-1", MAX_PARAM_STRING_LEN);

  Zoltan_Assign_Param_Vals(zz->Params, Color_params, zz->Debug_Level, zz->Proc,
			   zz->Debug_Proc);

  /* Check validity of parameters - they should be consistent with Zoltan_Color */
  if (!strcasecmp(coloring_problemStr, "distance-1"))
      coloring_problem = '1';
  else if (!strcasecmp(coloring_problemStr, "distance-2"))
      coloring_problem = '2';
  else if (!strcasecmp(coloring_problemStr, "partial-distance-2")
      || !strcasecmp(coloring_problemStr, "bipartite"))
      coloring_problem = 'P';
  else {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Unknown coloring requested. Using Distance-1 coloring.");
      coloring_problem = '1';
  }
  if (ss == 0) {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Invalid superstep size. Using default value 100.");
      ss = 100;
  }
  if (comm_pattern != 'S' && comm_pattern != 'A') {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Invalid communication pattern. Using synchronous communication (S).");
      comm_pattern = 'S';
  }
  if (comm_pattern == 'A' && (coloring_problem == '2' || coloring_problem == 'P')) {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Asynchronous communication pattern is not implemented for distance-2 coloring and its variants. Using synchronous communication (S).");
      comm_pattern = 'S';
  }
  if (coloring_order != 'I' && coloring_order != 'B' && coloring_order != 'U' && coloring_order != 'L' && coloring_order != 'N' && coloring_order != 'S') {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Invalid coloring order. Using internal first coloring order (I).");
      coloring_order = 'I';
  }
  if (coloring_order == 'U' && (coloring_problem == '2' || coloring_problem == 'P')) {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Interleaved coloring order is not implemented for distance-2 coloring and its variants. Using internal first coloring order (I).");
      coloring_order = 'I';
  }
  if (coloring_method !='F') {
      ZOLTAN_PRINT_WARN(zz->Proc, yo, "Invalid coloring method. Using first fit method (F).");
      coloring_method = 'F';
  }

  /* Compute Max number of array entries per ID over all processors.
     This is a sanity-maintaining step; we don't want different
     processors to have different values for these numbers. */
  comm[0] = zz->Num_GID;
  comm[1] = zz->Num_LID;
  MPI_Allreduce(comm, gcomm, 2, MPI_INT, MPI_MAX, zz->Communicator);
  zz->Num_GID = *num_gid_entries = gcomm[0];
  zz->Num_LID = *num_lid_entries = gcomm[1];

  /* Return if this processor is not in the Zoltan structure's
     communicator. */
  if (ZOLTAN_PROC_NOT_IN_COMMUNICATOR(zz))
      return ZOLTAN_OK;

  /* BUILD THE GRAPH */
  /* Check that the user has allocated space for the return args. */
  if (!color_exp)
      ZOLTAN_COLOR_ERROR(ZOLTAN_FATAL, "Color argument is NULL. Please give colors of local vertices.");


  requested_GNOs = (ZOLTAN_GNO_TYPE *) ZOLTAN_MALLOC(num_obj *
                                                     sizeof(ZOLTAN_GNO_TYPE));
  Zoltan_ZG_Build (zz, &graph, 0, 1, num_obj, 
                   global_ids, requested_GNOs);

  Zoltan_ZG_Export (zz, &graph,
		    &gvtx, &nvtx, NULL, NULL, 
                    &vtxdist, &xadj, &adjncy, &adjproc,
		    NULL, NULL);

  if (gvtx > (ZOLTAN_GNO_TYPE)INT_MAX){
      if (zz->Proc == 0){
          fprintf(stderr,
                  "Zoltan_Color_Test assumes number of vertices (%ld) is less than INT_MAX\n",gvtx);
      }
      ierr = ZOLTAN_FATAL;
      goto End;
  }

KDDKDDKDD(zz->Proc, "COLORTEST DD");
  /* Exchange global color information */
  if (vtxdist[zz->Num_Proc] && !(color = (int *) ZOLTAN_CALLOC(vtxdist[zz->Num_Proc], sizeof(int))))
      MEMORY_ERROR;

  if (nvtx && !(local_GNOs = (ZOLTAN_GNO_TYPE *) ZOLTAN_MALLOC(nvtx * sizeof(ZOLTAN_GNO_TYPE))))
      MEMORY_ERROR;
  for (i=0; i<nvtx; ++i)
      local_GNOs[i] = i+vtxdist[zz->Proc];
  if (vtxdist[zz->Num_Proc] && !(global_GNOs = (ZOLTAN_GNO_TYPE *) ZOLTAN_MALLOC(vtxdist[zz->Num_Proc] * sizeof(ZOLTAN_GNO_TYPE))))
      MEMORY_ERROR;
  for (i=0; i<vtxdist[zz->Num_Proc]; ++i)
      global_GNOs[i] = i;

  ierr = Zoltan_DD_Create (&dd_color, zz->Communicator, 
                           sizeof(ZOLTAN_GNO_TYPE)/sizeof(ZOLTAN_ID_TYPE), 0, 0, num_obj, 0);
  if (ierr != ZOLTAN_OK)
      ZOLTAN_COLOR_ERROR(ierr, "Cannot construct DDirectory.");
  /* Put req obs with 1 but first inialize the rest with 0 */
  ierr = Zoltan_DD_Update (dd_color, (ZOLTAN_ID_PTR)local_GNOs, NULL,
                           NULL, color, nvtx);
  if (ierr != ZOLTAN_OK)
      ZOLTAN_COLOR_ERROR(ierr, "Cannot update DDirectory.");
  ierr = Zoltan_DD_Update (dd_color, (ZOLTAN_ID_PTR)requested_GNOs, NULL,
                           NULL, color_exp, num_obj);
  if (ierr != ZOLTAN_OK)
      ZOLTAN_COLOR_ERROR(ierr, "Cannot update DDirectory.");
  /* Get requested colors from the DDirectory. */
  ierr = Zoltan_DD_Find (dd_color, (ZOLTAN_ID_PTR)global_GNOs, NULL, NULL,
                         color, vtxdist[zz->Num_Proc], NULL);

  if (ierr != ZOLTAN_OK)
      ZOLTAN_COLOR_ERROR(ierr, "Cannot find object in DDirectory.");
  /* Free DDirectory */
  Zoltan_DD_Destroy(&dd_color);
  ZOLTAN_FREE(&local_GNOs);
  ZOLTAN_FREE(&global_GNOs);

KDDKDDKDD(zz->Proc, "COLORTEST CHECK");

  if (coloring_problem == 'P' || coloring_problem == '2') {
      if (vtxdist[zz->Num_Proc] && !(partialD2 = (int *) ZOLTAN_CALLOC(vtxdist[zz->Num_Proc], sizeof(int))))
	  MEMORY_ERROR;
      if (vtxdist[zz->Num_Proc] && !(loc_partialD2 = (int *) ZOLTAN_CALLOC(vtxdist[zz->Num_Proc], sizeof(int))))
	  MEMORY_ERROR;

      if (coloring_problem == 'P') {
          for (i=0; i<num_obj; ++i) {
              int gno=requested_GNOs[i];
              loc_partialD2[gno] = 1;           
          }

          MPI_Allreduce(loc_partialD2, partialD2, vtxdist[zz->Num_Proc], MPI_INT, MPI_LOR, zz->Communicator);
      } else {
          for (i=0; i<vtxdist[zz->Num_Proc]; ++i)
              partialD2[i] = 1;
      }      
  }

  
  /* Check if there is an error in coloring */
  if (coloring_problem == '1') {
      for (i=0; i<nvtx; i++) {
          int gno = i + (int)vtxdist[zz->Proc];
          if (color[gno] <= 0) { /* object i is not colored properly */
              ierr = ZOLTAN_FATAL;
              printf("Error in coloring! u:%d, cu:%d\n", gno, color[gno]);               
              break;
          }
          for (j = xadj[i]; j < xadj[i+1]; ++j) {
              int v = (int)adjncy[j];
              if (color[gno] == color[v]) { /* neighbors have the same color */
                  ierr = ZOLTAN_FATAL;
                  printf("Error in coloring! u:%d, v:%d, cu:%d, cv:%d\n", gno, v, color[gno], color[v]); 
                  break; 
              }
          }
          if (ierr == ZOLTAN_FATAL)
              break;
      }
  } else if (coloring_problem == '2' || coloring_problem == 'P') {
      for (i=0; i<nvtx; i++) {
          int gno = i + (int)vtxdist[zz->Proc];
          if (partialD2[gno] && color[gno] <= 0) { /* object i is not colored properly */
              ierr = ZOLTAN_FATAL;
              printf("Error in coloring! u:%d, cu:%d\n", gno, color[gno]); 
              break;
          }
          for (j = xadj[i]; j < xadj[i+1]; ++j) {
              int v = (int)adjncy[j], k;
              if (partialD2[v] && color[v] <= 0) {
                  ierr = ZOLTAN_FATAL;
                  printf("Error in coloring! d1-neigh: u:%d, v:%d, cu:%d, cv:%d  pu:%d pv:%d\n", gno, v, color[gno], color[v], partialD2[gno], partialD2[v]);
                  break;
              }
              if (partialD2[gno] && partialD2[v] && color[gno] == color[v]) { /* d-1 neighbors have the same color */
                  ierr = ZOLTAN_FATAL;
                  printf("Error in coloring! d1-neigh: u:%d, v:%d, cu:%d, cv:%d  pu:%d pv:%d\n", gno, v, color[gno], color[v], partialD2[gno], partialD2[v]); 
                  break;
              }
              for (k = j+1; k < xadj[i+1]; ++k) {
                  int w = (int)adjncy[k];
                  if (partialD2[v] && partialD2[w] && color[v] == color[w]) { /* d-2 neighbors have the same color */
                      ierr = ZOLTAN_FATAL;
                      printf("Error in coloring! d2-neigh: v:%d, w:%d, cv:%d, cw:%d   pv:%d pw:%d\n", v, w, color[v], color[w], partialD2[v], partialD2[w]); 
                      break;

                  }

              }
          }
          if (ierr == ZOLTAN_FATAL)
              break;
      }      
  } else 
      ZOLTAN_COLOR_ERROR(ZOLTAN_WARN, "Zoltan_Color_Test is only implemented for distance-1 and distance-2 coloring. Unknown coloring, skipping verification.");      
  
 End:
KDDKDDKDD(zz->Proc, "COLORTEST DONE");
  if (ierr==ZOLTAN_FATAL)
      ierr = 2;
  MPI_Allreduce(&ierr, &ferr, 1, MPI_INT, MPI_MAX, zz->Communicator);
  if (ferr == 2)
      ierr = ZOLTAN_FATAL;
  else
      ierr = ZOLTAN_OK;

  Zoltan_ZG_Free (&graph);
  ZOLTAN_FREE(&adjproc);
  ZOLTAN_FREE(&color);
  ZOLTAN_FREE(&requested_GNOs);
  ZOLTAN_FREE(&partialD2);
  ZOLTAN_FREE(&loc_partialD2);

  return ierr;
}
Пример #3
0
/* if !copy, inmat is not usable after this call */
int
Zoltan_Matrix2d_Distribute (ZZ* zz, Zoltan_matrix inmat, /* Cannot be const as we can share it inside outmat */
			    Zoltan_matrix_2d *outmat, int copy)
{
  static char *yo = "Zoltan_Matrix2d_Distribute";
  int ierr = ZOLTAN_OK;
  int nProc_x, nProc_y;
  int myProc_x, myProc_y;
  int i, j, cnt;
  int *proclist = NULL;
  Zoltan_Arc *nonzeros= NULL, *sendbuf= NULL;
  ZOLTAN_GNO_TYPE *perm_y = NULL;
  float *wgtarray = NULL;
  float *tmpwgtarray = NULL;
  int msg_tag = 1021982;
  ZOLTAN_COMM_OBJ *plan;
  MPI_Comm communicator = MPI_COMM_NULL;
  int nProc;
  ZOLTAN_GNO_TYPE *yGNO = NULL;
  ZOLTAN_GNO_TYPE *pinGNO = NULL;
  ZOLTAN_GNO_TYPE tmp_gno;
  void *partdata = NULL;
  MPI_Datatype zoltan_gno_mpi_type;

  ZOLTAN_TRACE_ENTER(zz, yo);

  zoltan_gno_mpi_type = Zoltan_mpi_gno_type();

  memcpy(&outmat->mtx, &inmat, sizeof(Zoltan_matrix));
  if(copy) {
    /* TODO: We need to copy the arrays also */
    Zoltan_Matrix_Reset (&outmat->mtx);
    /* Copy also directories */
    outmat->mtx.ddX = Zoltan_DD_Copy (inmat.ddX);
    if (inmat.ddY == inmat.ddX)
      outmat->mtx.ddY = outmat->mtx.ddX;
    else
      outmat->mtx.ddY = Zoltan_DD_Copy (inmat.ddY);
  }

  communicator = outmat->comm->Communicator;
  nProc = outmat->comm->nProc;

  nProc_x = outmat->comm->nProc_x;
  nProc_y = outmat->comm->nProc_y;
  myProc_x = outmat->comm->myProc_x;
  myProc_y = outmat->comm->myProc_y;

KDDKDDKDD(zz->Proc, "    Zoltan_Matrix_Remove_Duplicates");
  ierr = Zoltan_Matrix_Remove_Duplicates(zz, outmat->mtx, &outmat->mtx);

/* KDDKDDKDD  FIX INDENTATION OF THIS BLOCK */
if (inmat.opts.speed != MATRIX_NO_REDIST) {
  if (outmat->hashDistFct == (distFnct *)&Zoltan_Distribute_Origin) {
    /* I need to know the original distribution */
    if (outmat->mtx.ddX != outmat->mtx.ddY) { /* No initial distribution */
      outmat->hashDistFct = (distFnct *)&Zoltan_Distribute_Linear;
    }
    else {
      int *cmember = NULL;

      cmember = (int*)ZOLTAN_MALLOC(outmat->mtx.nY*sizeof(int));
      if (outmat->mtx.nY > 0 && cmember == NULL) MEMORY_ERROR;
      Zoltan_DD_Find (outmat->mtx.ddY, (ZOLTAN_ID_PTR)outmat->mtx.yGNO, NULL, (char *)cmember, NULL,
		      outmat->mtx.nY, NULL);
KDDKDDKDD(zz->Proc, "    Zoltan_Distribute_Partition_Register");
      partdata = Zoltan_Distribute_Partition_Register(zz, outmat->mtx.nY, outmat->mtx.yGNO,
						      cmember, zz->Num_Proc, zz->Num_Proc);
      ZOLTAN_FREE(&cmember);
      Zoltan_Distribute_Set(outmat, (distFnct *)&Zoltan_Distribute_Origin, partdata);
    }
  }

  /*
   * Build comm plan for sending non-zeros to their target processors in
   * 2D data distribution.
   */
  /* TRICK: create fake arc (edgeno, -1) for empty Y. Upper bound for size might be nPins + nY */
  proclist = (int *)ZOLTAN_MALLOC((outmat->mtx.nPins+outmat->mtx.nY) *sizeof(int));
  sendbuf = (Zoltan_Arc*) ZOLTAN_MALLOC((outmat->mtx.nPins +outmat->mtx.nY)* sizeof(Zoltan_Arc));

  if ((outmat->mtx.nPins + outmat->mtx.nY >0) && (proclist == NULL || sendbuf == NULL)) MEMORY_ERROR;

  wgtarray = (float*) ZOLTAN_MALLOC((outmat->mtx.nPins+outmat->mtx.nY)*outmat->mtx.pinwgtdim*sizeof(float));

  if (outmat->mtx.nPins*outmat->mtx.pinwgtdim && !wgtarray) MEMORY_ERROR;

  yGNO = outmat->mtx.yGNO;
  pinGNO = outmat->mtx.pinGNO;

KDDKDDKDD(zz->Proc, "    CommPlan Hash");
  cnt = 0;
  for (i = 0; i < outmat->mtx.nY; i++) {
    ZOLTAN_GNO_TYPE edge_gno=-1;
    /* processor row for the edge */
    edge_gno = yGNO[i];

    for (j = outmat->mtx.ystart[i]; j < outmat->mtx.yend[i]; j++) {
      ZOLTAN_GNO_TYPE vtx_gno=-1;
      /* processor column for the vertex */
      vtx_gno = pinGNO[j];

      proclist[cnt] = (*outmat->hashDistFct)(edge_gno, vtx_gno, outmat->hashDistData,
					  &sendbuf[cnt].part_y);
      if (proclist[cnt] < 0) /* Discard this nnz */
        continue;
      sendbuf[cnt].GNO[0] = edge_gno;
      sendbuf[cnt].GNO[1] = vtx_gno;
      memcpy(wgtarray+cnt*outmat->mtx.pinwgtdim, outmat->mtx.pinwgt+j*outmat->mtx.pinwgtdim,
               outmat->mtx.pinwgtdim*sizeof(float));
      cnt++;
    }
    if(outmat->mtx.ystart[i] == outmat->mtx.yend[i]) {
      proclist[cnt] = (*outmat->hashDistFct)(edge_gno, -1, outmat->hashDistData,
					  &sendbuf[cnt].part_y);
      if (proclist[cnt] < 0) /* Discard this nnz */
        continue;
      sendbuf[cnt].GNO[0] = edge_gno;
      sendbuf[cnt].GNO[1] = -1;
      memset(wgtarray+cnt*outmat->mtx.pinwgtdim, 0,outmat->mtx.pinwgtdim*sizeof(float));
      cnt++;
    }
  }

  if (outmat->hashDistFct == (distFnct *)&Zoltan_Distribute_Origin)
    Zoltan_Distribute_Partition_Free(&outmat->hashDistData);

  if (outmat->mtx.yend != outmat->mtx.ystart + 1)
    ZOLTAN_FREE(&outmat->mtx.yend);
  outmat->mtx.yend = NULL;
  ZOLTAN_FREE(&outmat->mtx.ystart);
  ZOLTAN_FREE(&outmat->mtx.yGNO);
  ZOLTAN_FREE(&outmat->mtx.pinGNO);
  ZOLTAN_FREE(&outmat->mtx.pinwgt);
  ZOLTAN_FREE(&outmat->mtx.yGID);


  /*
   * Send pins to their target processors.
   * They become non-zeros in the 2D data distribution.
   */

KDDKDDKDD(zz->Proc, "    CommPlan Create");
  msg_tag--;
  ierr = Zoltan_Comm_Create(&plan, cnt, proclist, communicator, msg_tag, &outmat->mtx.nPins);
  ZOLTAN_FREE(&proclist);

  nonzeros = (Zoltan_Arc *) ZOLTAN_MALLOC((outmat->mtx.nPins) * sizeof(Zoltan_Arc));
  if (outmat->mtx.nPins && nonzeros == NULL) MEMORY_ERROR;

  msg_tag--;
  Zoltan_Comm_Do(plan, msg_tag, (char *) sendbuf, sizeof(Zoltan_Arc), (char *) nonzeros);
  ZOLTAN_FREE(&sendbuf);

  if (outmat->mtx.pinwgtdim) { /* We have to take care about weights */
    tmpwgtarray = (float*) ZOLTAN_MALLOC(outmat->mtx.nPins*outmat->mtx.pinwgtdim*sizeof(float));
    if (outmat->mtx.nPins && tmpwgtarray == NULL) MEMORY_ERROR;

    msg_tag--;
    Zoltan_Comm_Do(plan, msg_tag, (char *) wgtarray, outmat->mtx.pinwgtdim*sizeof(float),
		   (char *) tmpwgtarray);
    ZOLTAN_FREE(&wgtarray);
  }
  Zoltan_Comm_Destroy(&plan);

  /* Unpack the non-zeros received. */

KDDKDDKDD(zz->Proc, "    Zoltan_Matrix_Remove_DupArcs");
  /* TODO: do take care about singletons */
  Zoltan_Matrix_Remove_DupArcs(zz, outmat->mtx.nPins, (Zoltan_Arc*)nonzeros, tmpwgtarray,
			       &outmat->mtx);
}

  /* Now we just have to change numbering */
  outmat->dist_y = (ZOLTAN_GNO_TYPE *) ZOLTAN_CALLOC((nProc_y+1), sizeof(ZOLTAN_GNO_TYPE));
  outmat->dist_x = (ZOLTAN_GNO_TYPE *) ZOLTAN_CALLOC((nProc_x+1), sizeof(ZOLTAN_GNO_TYPE));
  if (outmat->dist_y == NULL || outmat->dist_x == NULL) MEMORY_ERROR;

  /* FIXME: Work only in 1D */
  tmp_gno = (ZOLTAN_GNO_TYPE)outmat->mtx.nY;
  MPI_Allgather(&tmp_gno, 1, zoltan_gno_mpi_type, outmat->dist_y+1, 1, zoltan_gno_mpi_type, communicator);
  for (i = 1 ; i <= nProc_y ; i ++) {
    outmat->dist_y[i] += outmat->dist_y[i-1];
  }
  outmat->dist_x[1] = outmat->mtx.globalX;

  perm_y = (ZOLTAN_GNO_TYPE *) ZOLTAN_MALLOC(outmat->mtx.nY * sizeof(ZOLTAN_GNO_TYPE));
  if (outmat->mtx.nY > 0 && perm_y == NULL) MEMORY_ERROR;
  for (i = 0 ; i < outmat->mtx.nY ; ++i){
    perm_y[i] = i + outmat->dist_y[myProc_y];
  }

KDDKDDKDD(zz->Proc, "    Zoltan_Matrix_Permute");
  Zoltan_Matrix_Permute(zz, &outmat->mtx, perm_y);

KDDKDDKDD(zz->Proc, "    Zoltan_Matrix_Permute done");
 End:
  ZOLTAN_FREE(&perm_y);
  ZOLTAN_FREE(&proclist);
  ZOLTAN_FREE(&sendbuf);
  ZOLTAN_FREE(&nonzeros);
  ZOLTAN_FREE(&tmpwgtarray);


  ZOLTAN_TRACE_EXIT(zz, yo);

  return (ierr);
}