compiling with MPI, METIS, CGNS on iMac
 

Hi,

I am trying to compile on my mac. I have downloaded and compiled openmpi and cgns with success, but I had a lot of warnings with metis - I used v4.0.3 as suggested elsewhere on here.

My configure command for SU2 is as specified on the website, with the output from make below. Looks like problems with the metis libs?

Any help much appreciated.

Thanks.

Making all in trunk/Common/lib
/usr/local/bin/mpicxx -DPACKAGE_NAME=\"SU\^2\" -DPACKAGE_TARNAME=\"SU\^2\" -DPACKAGE_VERSION=\"2.0\" -DPACKAGE_STRING=\"SU\^2\ 2.0\" -DPACKAGE_BUGREPORT=\"susquared-dev@lists.stanford.edu\" -DPACKAGE=\"SU\^2\" -DVERSION=\"2.0\" -I. -I/Users/Simon/CFD/cgnslib_3.1.4/src/ -I/Users/Simon/CFD/metis-4.0.3/Lib -DMETIS_5 -DNO_TECIO -g -O2 -MT ../src/libSU2_a-config_structure.o -MD -MP -MF ../src/.deps/libSU2_a-config_structure.Tpo -c -o ../src/libSU2_a-config_structure.o `test -f '../src/config_structure.cpp' || echo './'`../src/config_structure.cpp
mv -f ../src/.deps/libSU2_a-config_structure.Tpo ../src/.deps/libSU2_a-config_structure.Po
/usr/local/bin/mpicxx -DPACKAGE_NAME=\"SU\^2\" -DPACKAGE_TARNAME=\"SU\^2\" -DPACKAGE_VERSION=\"2.0\" -DPACKAGE_STRING=\"SU\^2\ 2.0\" -DPACKAGE_BUGREPORT=\"susquared-dev@lists.stanford.edu\" -DPACKAGE=\"SU\^2\" -DVERSION=\"2.0\" -I. -I/Users/Simon/CFD/cgnslib_3.1.4/src/ -I/Users/Simon/CFD/metis-4.0.3/Lib -DMETIS_5 -DNO_TECIO -g -O2 -MT ../src/libSU2_a-dual_grid_structure.o -MD -MP -MF ../src/.deps/libSU2_a-dual_grid_structure.Tpo -c -o ../src/libSU2_a-dual_grid_structure.o `test -f '../src/dual_grid_structure.cpp' || echo './'`../src/dual_grid_structure.cpp
mv -f ../src/.deps/libSU2_a-dual_grid_structure.Tpo ../src/.deps/libSU2_a-dual_grid_structure.Po
/usr/local/bin/mpicxx -DPACKAGE_NAME=\"SU\^2\" -DPACKAGE_TARNAME=\"SU\^2\" -DPACKAGE_VERSION=\"2.0\" -DPACKAGE_STRING=\"SU\^2\ 2.0\" -DPACKAGE_BUGREPORT=\"susquared-dev@lists.stanford.edu\" -DPACKAGE=\"SU\^2\" -DVERSION=\"2.0\" -I. -I/Users/Simon/CFD/cgnslib_3.1.4/src/ -I/Users/Simon/CFD/metis-4.0.3/Lib -DMETIS_5 -DNO_TECIO -g -O2 -MT ../src/libSU2_a-geometry_structure.o -MD -MP -MF ../src/.deps/libSU2_a-geometry_structure.Tpo -c -o ../src/libSU2_a-geometry_structure.o `test -f '../src/geometry_structure.cpp' || echo './'`../src/geometry_structure.cpp
../src/geometry_structure.cpp: In constructor ‘CPhysicalGeometry::CPhysicalGeometry(CConfig*, std::string, short unsigned int, short unsigned int, short unsigned int)’:
../src/geometry_structure.cpp:1131: warning: format ‘%8i’ expects type ‘int’, but argument 3 has type ‘long int’
../src/geometry_structure.cpp:1131: warning: format ‘%8i’ expects type ‘int’, but argument 3 has type ‘long int’
../src/geometry_structure.cpp:1183: warning: format ‘%8i’ expects type ‘int’, but argument 3 has type ‘long int’
../src/geometry_structure.cpp:1183: warning: format ‘%8i’ expects type ‘int’, but argument 3 has type ‘long int’
../src/geometry_structure.cpp: In member function ‘virtual void CPhysicalGeometry::SetColorGrid(CConfig*)’:
../src/geometry_structure.cpp:4297: error: ‘METIS_NOPTIONS’ was not declared in this scope
../src/geometry_structure.cpp:4298: error: ‘options’ was not declared in this scope
../src/geometry_structure.cpp:4298: error: ‘METIS_SetDefaultOptions’ was not declared in this scope
../src/geometry_structure.cpp:4299: error: ‘METIS_OPTION_OBJTYPE’ was not declared in this scope
../src/geometry_structure.cpp:4299: error: ‘METIS_OBJTYPE_CUT’ was not declared in this scope
/Users/Simon/CFD/metis-4.0.3/Lib/proto.h:233: error: too many arguments to function ‘void METIS_PartMeshNodal(int*, int*, idxtype*, int*, int*, int*, int*, idxtype*, idxtype*)’
../src/geometry_structure.cpp:4300: error: at this point in file
make[1]: *** [../src/libSU2_a-geometry_structure.o] Error 1
make: *** [all-recursive] Error 1

Probably you are setting --with-Metis-version=5 in you configure command for SU2, but as far I have seen you are using metis 4.

Could you please get rid of --with-Metis-version=5 in the configure command?

Cheers,
Francisco

Hi Francisco,

Thanks, removing --with-Metis-version=5 has done the trick. I forgot I had left it in the ./configure command having switched from metis v5 to v4.

I've run the oneram6 tutorial following the parallel computation approach. I ran the parallel_computation.py script in /usr/local/bin where the SU2 exes are but it couldnt find the DDC or SU2 exes. Looking at the python script it was calling to $SU2_HOME/SU2Py which doesnt contain any of teh SU2 exes (I have set my $SU2_HOME to the trunk folder as specified). I modified the python script to just call SU2_DDC and CFD (calling from /usr/local/bin) and that has worked fine. I'm also curious though as there are also SU2_ exes in $SU2_HOME?

I also have a question on memory usage - running the oneram6 example in serial the memory usage is ~1150MB but in parallel on 4 cores it is ~450MB per core which is almost twice as much in total (4*450=1800MB)?

Thanks for your help,

Simon.

$su2_run
 

Have you gone to the directory to confirm the location of the python scripts?

It depends on which release you are using, but I noticed a change in the file structure when I started following the developer releases. For instance, my SU2_RUN stuff (duly added to my path) is all located under usr/local/bin...but my SU2_HOME is just pointing to SU2_v2.0.3...ie: no trunk anymore.

I am using v2.0. It seems there are python scripts in SU2_HOME/trunk/SU2Py and also in SU2_RUN (/usr/local/bin) with the compiled exes. There are also exes in SU2_HOME/trunk dated 8th Jan so I assume they are binaries compiled at the v2.0 release...

Does anyone have any guidance on memory usage? A rule of thumb for no elements vs RAM would be really handy..

It seems for running the implicit viscose flow, per 1 million cells you probably need about 1 gig RAM, also it assume that you are using single precision. if you use double precision, you will need twice, it means per 1 million cells you will need 2 gig RAM. I am speaking from experience, it may be different from one code to another but still it can be used as a rule of thumb. for example for ONERA M6 Wing if you use 5 million cells in double precision you will need something around 12 gig RAM.( it's a little bit more than twice) and if you use 300K cells you will need 1 gig RAM if you are using double precision.