 |
 |
 |
|
|
|
[Sponsors] | ||||
February 14, 2012, 04:16
|
Metis for OpenFOAM 2.1.0
|
#1 |
|
Senior Member
Onno
Join Date: Jan 2012
Location: Germany
Posts: 120
Rep Power: 15  |
Hello,
the official support for metis has been dropped for license issues. Has someone been able to install and use metis on their own? My attempts so far have been fruitless. Greetings Kaskade |
|
|
|
|
|
June 5, 2012, 18:17
|
|
#2 |
|
Senior Member
Robert Sawko
Join Date: Mar 2009
Posts: 117
Rep Power: 22 |
Yes, you need to download metis from this site
http://people.sc.fsu.edu/~jburkardt/c_src/metis/metis.html and GKLib. GKLib was a bit more tricky to find but I have found it for example here: http://code.google.com/p/graphlabapi...0a9d384317c3bf You can checkout the whole repository if you have mercurial installed. Now it's the question of compiling these things. Set up your include paths correctly. For GKLib I had to add Code:
DMALLOCINC = -I../libmetis/ Metis compiled seamlessly in my case using script although you must add -fPIC to the gcc command. Finally go to Code:
$FOAM_SRC/parallel/decompose/metisDecomp Code:
wmake libso EDIT: ALSO! I removed the -I pointing to the dummy library. I guess that might be important. :-) Last edited by AlmostSurelyRob; June 5, 2012 at 18:34. |
|
|
|
|
|
|
June 23, 2012, 06:31
|
|
#3 |
|
Senior Member
Onno
Join Date: Jan 2012
Location: Germany
Posts: 120
Rep Power: 15 |
Thanks, I'll try it some time. My current case is pretty symmetric, so simple does the trick.
|
|
|
|
|
|
|
April 9, 2013, 07:39
|
|
#4 |
|
Member
pere
Join Date: Mar 2011
Location: barcelona
Posts: 46
Rep Power: 15 |
Hi everybody,
I'm trying to compile metisDecomp, I'm getting this error while I'm doing wmake libso: metisDecomp.C: In member function ‘Foam::label Foam::metisDecomp::decompose(const Foam::List<int>&, const Foam::List<int>&, const Foam::scalarField&, Foam::List<int>&)’: metisDecomp.C:237: error: ‘METIS_WPartGraphRecursive’ was not declared in this scope metisDecomp.C:254: error: cannot convert ‘int*’ to ‘real_t*’ for argument ‘9’ to ‘int METIS_PartGraphRecursive(idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, real_t*, real_t*, idx_t*, idx_t*, idx_t*)’ metisDecomp.C:275: error: ‘METIS_WPartGraphKway’ was not declared in this scope metisDecomp.C:292: error: cannot convert ‘int*’ to ‘real_t*’ for argument ‘9’ to ‘int METIS_PartGraphKway(idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, idx_t*, real_t*, real_t*, idx_t*, idx_t*, idx_t*)’ make: *** [Make/linux64GccDPOpt/metisDecomp.o] Error 1 I've linked to libmetis.so.... perhaps any include that is not found? |
|
|
|
|
|
|
April 9, 2013, 12:20
|
|
#5 |
|
Senior Member
Onno
Join Date: Jan 2012
Location: Germany
Posts: 120
Rep Power: 15 |
I've compared scotch and metis for some of my old meshes. For some metis was a little faster (decomposing and/or simulation), for some scotch. You are probably wasting more time getting metis to run, than you would ever save by using it.
|
|
|
|
|
|
|
April 9, 2013, 12:40
|
|
#6 |
|
Member
pere
Join Date: Mar 2011
Location: barcelona
Posts: 46
Rep Power: 15 |
Hi Kaskade,
I'm compiling OpenFoam from the source code so I need the same time to compile it with scotch or metis. Thanks for your reply. |
|
|
|
|
|
|
April 9, 2013, 17:30
|
|
#7 |
|
Member
Dan Kokron
Join Date: Dec 2012
Posts: 33
Rep Power: 14 |
Mods to get 2.2.x to compile with METIS from parmetis-4.0.2.
Make sure you test the results!! Code:
git diff src/parallel/decompose/Allwmake
diff --git a/src/parallel/decompose/Allwmake b/src/parallel/decompose/Allwmake
index 5cea1fb..581b877 100755
--- a/src/parallel/decompose/Allwmake
+++ b/src/parallel/decompose/Allwmake
@@ -62,6 +62,7 @@ else
echo
fi
+wmake $makeType metisDecomp
wmake $makeType decompositionMethods
wmake $makeType decompose
Code:
git diff src/parallel/decompose/metisDecomp/Make/options
diff --git a/src/parallel/decompose/metisDecomp/Make/options b/src/parallel/decompose/metisDecomp/Make/options
index a6a0f14..b046053 100644
--- a/src/parallel/decompose/metisDecomp/Make/options
+++ b/src/parallel/decompose/metisDecomp/Make/options
@@ -1,7 +1,6 @@
EXE_INC = \
- -I$(WM_THIRD_PARTY_DIR)/metis-5.0pre2/include \
- -I../../dummyThirdParty/metisDecomp /* location of dummy metis.h */ \
+ -I/home/dkokron/play/ParMETIS/parmetis-4.0.2/install/include \
-I../decompositionMethods/lnInclude
LIB_LIBS = \
- -L$(FOAM_EXT_LIBBIN) -lmetis -lGKlib
+ -L$(FOAM_EXT_LIBBIN) -lmetis
Code:
git diff src/parallel/decompose/metisDecomp/metisDecomp.C
diff --git a/src/parallel/decompose/metisDecomp/metisDecomp.C b/src/parallel/decompose/metisDecomp/metisDecomp.C
index e1614a8..2530e61 100644
--- a/src/parallel/decompose/metisDecomp/metisDecomp.C
+++ b/src/parallel/decompose/metisDecomp/metisDecomp.C
@@ -202,6 +202,7 @@ Foam::label Foam::metisDecomp::decompose
// Vertex weight info
int wgtFlag = 0;
+ int ncon = 1;
int* vwgtPtr = NULL;
int* adjwgtPtr = NULL;
@@ -220,18 +221,19 @@ Foam::label Foam::metisDecomp::decompose
{
if (processorWeights.size())
{
- METIS_WPartGraphRecursive
+ METIS_PartGraphRecursive
(
&numCells, // num vertices in graph
+ &ncon, // number of balancing constraints. It should be at least 1.
const_cast<List<int>&>(xadj).begin(), // indexing into adjncy
const_cast<List<int>&>(adjncy).begin(), // neighbour info
vwgtPtr, // vertexweights
+ NULL,
adjwgtPtr, // no edgeweights
- &wgtFlag,
- &numFlag,
&nProcs,
- processorWeights.begin(),
- options.begin(),
+ NULL, // NULL indicates graph should be equally divided among the partitions
+ NULL, // load imbalance tolerance for each constraint; NULL -> 1.001
+ NULL, // options array
&edgeCut,
finalDecomp.begin()
);
@@ -241,14 +243,16 @@ Foam::label Foam::metisDecomp::decompose
METIS_PartGraphRecursive
(
&numCells, // num vertices in graph
+ &ncon, // number of balancing constraints. It should be at least 1.
const_cast<List<int>&>(xadj).begin(), // indexing into adjncy
const_cast<List<int>&>(adjncy).begin(), // neighbour info
vwgtPtr, // vertexweights
+ NULL,
adjwgtPtr, // no edgeweights
- &wgtFlag,
- &numFlag,
&nProcs,
- options.begin(),
+ NULL, // NULL indicates graph should be equally divided among the partitions
+ NULL, // load imbalance tolerance for each constraint; NULL -> 1.001
+ NULL, // options array
&edgeCut,
finalDecomp.begin()
);
@@ -258,18 +262,19 @@ Foam::label Foam::metisDecomp::decompose
{
if (processorWeights.size())
{
- METIS_WPartGraphKway
+ METIS_PartGraphKway
(
&numCells, // num vertices in graph
+ &ncon, // number of balancing constraints. It should be at least 1.
const_cast<List<int>&>(xadj).begin(), // indexing into adjncy
const_cast<List<int>&>(adjncy).begin(), // neighbour info
vwgtPtr, // vertexweights
+ NULL,
adjwgtPtr, // no edgeweights
- &wgtFlag,
- &numFlag,
&nProcs,
- processorWeights.begin(),
- options.begin(),
+ NULL, // NULL indicates graph should be equally divided among the partitions
+ NULL, // load imbalance tolerance for each constraint; NULL -> 1.001
+ NULL, // options array
&edgeCut,
finalDecomp.begin()
);
@@ -279,14 +284,16 @@ Foam::label Foam::metisDecomp::decompose
METIS_PartGraphKway
(
&numCells, // num vertices in graph
+ &ncon, // number of balancing constraints. It should be at least 1.
const_cast<List<int>&>(xadj).begin(), // indexing into adjncy
const_cast<List<int>&>(adjncy).begin(), // neighbour info
vwgtPtr, // vertexweights
+ NULL,
adjwgtPtr, // no edgeweights
- &wgtFlag,
- &numFlag,
&nProcs,
- options.begin(),
+ NULL, // NULL indicates graph should be equally divided among the partitions
+ NULL, // load imbalance tolerance for each constraint; NULL -> 1.001
+ NULL, // options array
&edgeCut,
finalDecomp.begin()
);
@@ -328,7 +335,16 @@ Foam::labelList Foam::metisDecomp::decompose
}
CompactListList<label> cellCells;
- calcCellCells(mesh, identity(mesh.nCells()), mesh.nCells(), cellCells);
+ calcCellCells
+ (
+ mesh,
+ identity(mesh.nCells()),
+ mesh.nCells(),
+ true,
+ cellCells
+ );
+
+ // DSK replace after switching to 2.2.x calcCellCells(mesh, identity(mesh.nCells()), mesh.nCells(), cellCells);
// Decompose using default weights
labelList decomp;
@@ -362,7 +378,15 @@ Foam::labelList Foam::metisDecomp::decompose
// xadj(celli) : start of information in adjncy for celli
CompactListList<label> cellCells;
- calcCellCells(mesh, agglom, agglomPoints.size(), cellCells);
+ calcCellCells
+ (
+ mesh,
+ agglom,
+ agglomPoints.size(),
+ true,
+ cellCells
+ );
+ // DSK replace after switching to 2.2.x calcCellCells(mesh, agglom, agglomPoints.size(), cellCells);
// Decompose using default weights
labelList finalDecomp;
|
|
|
|
|
|
|
April 10, 2013, 06:09
|
|
#8 |
|
Member
pere
Join Date: Mar 2011
Location: barcelona
Posts: 46
Rep Power: 15 |
Thanks I will try it!
|
|
|
|
|
|
|
May 31, 2014, 02:20
|
Hi pere ,did you solve this problem?
|
#9 | |
|
New Member
Ruby Qian
Join Date: Aug 2013
Location: Nanjing,Jiangsu,PRC
Posts: 13
Rep Power: 13 |
Quote:
|
||
|
|
|
||
 |
|
|
Similar Threads
|
||||
| Thread | Thread Starter | Forum | Replies | Last Post |
| metis decompose OpenFoam 2.0.0 | sebastianweiper | OpenFOAM Installation | 11 | December 29, 2020 12:59 |
| How to partition EDGES by using METIS | Lupocci | Main CFD Forum | 0 | August 18, 2010 18:43 |
| Problem with Partitioners (METIS) | Partition | ANSYS Meshing & Geometry | 1 | January 1, 2010 14:53 |
| Node numbering at interface in MeTiS. | mehul | FLUENT | 0 | December 19, 2006 11:22 |
| Node numbering at interface in MeTiS. | mehul | Main CFD Forum | 0 | December 19, 2006 11:21 |
2Likes
Linear Mode
