mirror of
https://develop.openfoam.com/Development/openfoam.git
synced 2025-11-28 03:28:01 +00:00
156 lines
5.8 KiB
Plaintext
156 lines
5.8 KiB
Plaintext
|
|
|
|
METIS 4.0.2, 3/10/04
|
|
------------------------------------------------------------------------------
|
|
- Fixed a problem with weighted graphs and ometis.c
|
|
|
|
|
|
METIS 4.0.1, 11/29/98
|
|
------------------------------------------------------------------------------
|
|
This is mostly a bug-fix release
|
|
|
|
- Fixed some bugs in the multi-constraint partitioning routines
|
|
- Fixed some bugs in the volume-minimization routines
|
|
|
|
|
|
|
|
METIS 4.0.0, 9/20/98
|
|
------------------------------------------------------------------------------
|
|
METIS 4.0 contains a number of changes over the previous major release (ver
|
|
3.0.x). Most of these changes are concentrated on the graph and mesh
|
|
partitioning routines and they do not affect the sparse matrix re-ordering
|
|
routines. Here is a list of the major changes:
|
|
|
|
Multi-Constraint Partitioning
|
|
-----------------------------
|
|
METIS now includes partitioning routines that can be used to a partition
|
|
a graph in the presence of multiple balancing constraints.
|
|
|
|
Minimizing the Total Communication Volume
|
|
-----------------------------------------
|
|
METIS now includes partitioning routines whose objective is to minimize
|
|
the total communication volume (as opposed to minimizing the edge-cut).
|
|
|
|
Minimizing the Maximum Connectivity of the Subdomains
|
|
-----------------------------------------------------
|
|
The k-way partitioning routines in METIS can now directly minimize the number
|
|
of adjacent subdomains. For most graphs corresponding to finite element
|
|
meshes, METIS is able to significantly reduce the maximum (and total) number of
|
|
adjacent subdomains.
|
|
|
|
|
|
|
|
|
|
METIS 3.0.6, 1/28/98
|
|
-------------------------------------------------------------------------------
|
|
- Fixed some problems when too many partitions were asked, and each partition
|
|
end up having 0 vertices
|
|
- Fixed some bugs in the I/O routines
|
|
- Added support for the g77 compiler under Linux
|
|
|
|
|
|
METIS 3.0.5, 12/22/97
|
|
-------------------------------------------------------------------------------
|
|
- Fixed problems on 64-bit architectures (eg., -64 option on SGIs).
|
|
- Added some options in Makefile.in
|
|
|
|
|
|
METIS 3.0.4, 12/1/97
|
|
-------------------------------------------------------------------------------
|
|
Fixed a memory leak in the ordering code.
|
|
|
|
|
|
METIS 3.0.3, 11/5/97
|
|
-------------------------------------------------------------------------------
|
|
This is mostly a bug-fix release with just a few additions
|
|
|
|
Added functionality
|
|
- Added support for quadrilateral elements.
|
|
- Added a routine METIS_EstimateMemory that estimates the amount of
|
|
memory that will be allocated by METIS. This is useful in determining
|
|
if a problem can run on your system.
|
|
- Added hooks to allow PARMETIS to use the orderings produced by METIS.
|
|
This is hidden from the user but it will be used in the next release
|
|
of PARMETIS.
|
|
|
|
Bug-fixes
|
|
- Fixed a bug related to memory allocation. This should somewhat reduce the
|
|
overall memory used by METIS.
|
|
- Fixed some bugs in the 'graphchk' program in the case of weighted graphs.
|
|
- Removed some code corresponding to unused options.
|
|
- Fixed some minor bugs in the node-refinement code
|
|
|
|
|
|
|
|
-------------------------------------------------------------------------------
|
|
METIS 3.0 contains a number of changes over METIS 2.0.
|
|
The major changes are the following:
|
|
|
|
General Changes
|
|
---------------
|
|
1. Added code to directly partition finite element meshes.
|
|
|
|
2. Added code to convert finite element meshes into graphs so they
|
|
can be used by METIS.
|
|
|
|
1. The names, calling sequences, and options of the routines in
|
|
METISlib have been changed.
|
|
|
|
2. Better support has been added for Fortran programs.
|
|
|
|
3. Eliminated the 'metis' program. The only way to tune METIS's
|
|
behavior is to use METISlib.
|
|
|
|
4. Improved memory management. METIS should now only abort if truly
|
|
there is no more memory left in the system.
|
|
|
|
|
|
Graph Partitioning
|
|
------------------
|
|
1. Added partitioning routines that can be used to compute a partition
|
|
with prescribed partition weights. For example, they can be used to
|
|
compute a 3-way partition such that partition 1 has 50% of the weight,
|
|
partition 2 has 20% of the way, and partition 3 has 30% of the weight.
|
|
|
|
2. Improved the speed of the k-way partitioning algorithm (kmetis). The
|
|
new code has better cache locality which dramatically improves the
|
|
speed for large graphs. A factor of 4 speedup can be obtained for
|
|
certain graphs. METIS can now partition a 4 million node graph
|
|
in well under a minute on a MIPS R10000.
|
|
|
|
3. Eliminated some of the options that were seldom used.
|
|
|
|
|
|
Fill-Reducing Orderings
|
|
----------------------
|
|
1. Added a node based ordering code `onmetis' that greatly improves
|
|
ordering quality.
|
|
|
|
2. Improved the quality of the orderings produced by the original
|
|
edge-based ordering code (it is now called 'oemetis').
|
|
|
|
3. METIS can now analyze the graph and try to compress together
|
|
nodes with identical sparsity pattern. For some problems, this
|
|
significantly reduces ordering time
|
|
|
|
4. METIS can now prune dense columns prior to ordering. This can be
|
|
helpful for LP matrices.
|
|
|
|
|
|
Mesh Partitioning
|
|
-----------------
|
|
1. METIS can now directly partition the element node array of finite
|
|
element meshes. It produces two partitioning vectors. One for the
|
|
elements and one for the nodes. METIS supports the following
|
|
elements: triangles, tetrahedra, hexahedra
|
|
|
|
|
|
Mesh-To-Graph Conversion Routines
|
|
---------------------------------
|
|
1. METIS now includes a number of mesh conversion functions that can
|
|
be used to create the dual and nodal graphs directly from the
|
|
element connectivity arrays. These are highly optimized routines.
|
|
|
|
|
|
|