You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
lib
96dbec27e0
|
2 years ago | |
---|---|---|
.. | ||
README | 2 years ago | |
hier.c | 2 years ago | |
hier.h | 2 years ago | |
hier_free_struct.c | 2 years ago |
README
# @HEADER
#
########################################################################
#
# Zoltan Toolkit for Load-balancing, Partitioning, Ordering and Coloring
# Copyright 2012 Sandia Corporation
#
# Under the terms of Contract DE-AC04-94AL85000 with Sandia Corporation,
# the U.S. Government retains certain rights in this software.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are
# met:
#
# 1. Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
#
# 2. Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
#
# 3. Neither the name of the Corporation nor the names of the
# contributors may be used to endorse or promote products derived from
# this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
# LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
# NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
# Questions? Contact Karen Devine kddevin@sandia.gov
# Erik Boman egboman@sandia.gov
#
########################################################################
#
# @HEADER
HIER DIRECTORY -- Hierarchical partitioning and rebalancing support
-------------------------------------------------------------
README:
Text giving a brief description of the files located in this directory
hier.c:
Main implementation of hierarchical balancing
hier.h:
Structure, constant, macros, prototype definitions for hierarchical
balancing
hier_free_structure.c:
callback to free memory allocated for hierarchical balancing
Description of hierarchical balancing intermediate structure
------------------------------------------------------------
The intermediate structure allows hierarchical balancing to pass
around a small representation of the input during the stages of
balancing to avoid application migration until the end of the
process.
The following members of the struct HierPartParamsStruct are
involved in the storage of the intermediate structure:
int init_num_obj; /* number of local objects at start */
int hier_num_obj; /* number of local objects during
hierarchical balancing */
ZOLTAN_ID_PTR local_ids; /* / lists of local ids and global ids */
ZOLTAN_ID_PTR global_ids; /* \ for initial partitioning */
float *vwgt; /* vector of vertex weights */
int *input_parts; /* Initial partitions for objects. */
char *migrated; /* array of flags indicating whether
each gid in global_ids has been
migrated somewhere else */
int obj_wgt_dim, edge_wgt_dim; /* object and edge weight dimensions */
idxtype *vtxdist, *xadj; /* intermediate graph structure storage */
ZOLTAN_ID_PTR adjncy; /* see Zoltan_Build_Graph */
float *ewgts; /* edge weights for intermediate struct */
int ndims; /* number of dimensions for geom data */
int num_edges; /* number of edges in graph rep */
double *geom_vec; /* geometry of objects in intermediate */
char *migrated; /* array of flags indicating whether
each gid in global_ids has been
migrated somewhere else */
int num_migrated_in_gids; /* number of gids migrated to this proc */
int alloc_migrated_in_gids; /* size of allocated array of migrated
in gids */
ZOLTAN_ID_PTR migrated_in_gids; /* ordered array of gids migrated in */
void **migrated_in_data; /* data migrated in, parallel array to
migrated_in_gids */
struct Zoltan_DD_Struct *dd; /* distributed data to track migrated
gids during hierarchical balancing */
obj_wgt_dim and edge_wgt_dim are inherited from the main Zoltan_Struct
used by the application.
The structure is initially constructed by populating the init_num_obj,
global_ids, local_ids, vwgt, and input_parts members by a call to
Zoltan_Get_Obj_List. hier_num_obj is initialized to init_num_obj.
local_ids and global_ids contain the LIDs and GIDs of the objects that
are starting on the local process.
vwgt contains an array of weights for the objects on the local
process. vwgt[i*obj_wgt_dim]...vwgt[(i+1)*obj_wgt_dim-1] are the
weights for local object i.
input_parts is an array of the starting partition assignment of each
object that starts on the local process.
The members vtxdist, xadj, adjncy, and ewgts are filled by a call to
Zoltan_Build_Graph. If graph callbacks are not being used, only
vtxdist is populated. This builds what is usually used as input to
Parmetis or Jostle routines.
vtxdist is a globally-replicated array of size equal to the number of
processes+1, containing essentially a prefix sum of the number of
objects per process. If process 0 has 10 objects, process 1 has 7
objects, and process 2 has 8 objects, each process will have a vtxdist
array with the values {0,10,17,25}. These can be thought of as
starting indices on each process into a distributed global_ids array.
xadj is an array with one entry for each local object that indicates
the starting index into the adjncy array for the graph edges are
incident on that object.
adjncy is an array of graph edges. Local object i's incident edges
are listed in positions xadj[i] to xadj[i+1]. The values listed are
from the global numbering of objects as defined by adding the object's
position in a global_ids array and the vtxdist value for that process.
ewgts is the array of edge weights corresponding to the adjncy array,
with edge_wgt_dim entries per edge.
If geometric information is to be used, a call to
Zoltan_Get_Coordinates populates the ndims and geom_vec members.
ndims is the number of values in each coordinate and geom_vec contains
the actual coordinates for each object in global_ids.
Objects that have been migrated from their original locations are
represented in some additional members. The members described above
that describe the original object locations and information do not
change during the hierarchical balancing process.
Objects that have migrated from their original location are marked by
placing a 1 at the corresponding position in the "migrated" array on
their starting process.
Objects that have migrated to a process are maintained in the arrays
migrated_in_gids and migrated_in_data. The number of objects in these
arrays is stored in num_migrated_in_gids and the allocated size of
these arrays is stored in alloc_migrated_in_gids. A distributed data
directory that maintains current process assignments is stored in dd.
This allows all processes to know the location of an object when one
if its objects' edges needs this information.
migrated_in_gids does not contain the original application global ids,
but rather the integer that specifies its position in the original
global id arrays, computed by its position in its process's local
array and the offset specified in vtxdist.
migrated_in_gids is maintained in order to allow a binary search to be
used to locate a gid or to determine that it is not in the array
quickly.
The migrated_in_data array is maintained in the same order. Each
entry in migrated_in_data is a pointer to a struct HierGIDInfo,
defined as:
struct HierGIDInfo {
float *wgts;
double *coords;
int num_adj;
ZOLTAN_ID_PTR adj;
};
This structure stores an object's weights (object and edge),
coordinates, and edges. The wgts array has obj_wgt_dim entries for
object weights followed by edge_wgt_dim edge weights for each adjacent
edge. The coords array has ndims entries for geometric coordinates.
The adj array has gid numbers for the other object for each of the
num_adj edges.