Re: Compilation problem with mumps

[pierre.gosselet]

Dear Kostas, dear Edgar,

thank you for your help, I finally managed to compile.


For the record:
+ I had to use $LD_LIBRARY_PATH to give the path to mumps/scotch/metis
library (which I had to compile on my own with fpic), for some reason
configure failed if I tried to give the libraries some other way.

+ then the following ./configure found everything 

./configure CC=icc CXX=icpc FC=ifort --with-pic --disable-matlab 
--enable-python CXXFLAGS="-I/pathtomumpsheaders"

+ for the compilation, I had to remove the -ansi option which I think
makes icc refuse c++-style comments (//) which appear in superlu and
python headers.
+ For the final linking of the python interface, I had to use ifort
instead of icc and to add lots of libraries (*mumps*, scotch*, ...).

best regards,
pierre

Le mardi 25 mai 2021 à 22:24 +0200, Konstantinos Poulios a écrit :
> Dear Pierre
> 
> This is how I compile sequential getfem on our cluster (running some
> version of Scientific Linux):
> 
> I put all dependencies (symbolic links actually) into two folders:
> /some_path_with_dependencies/lib
> /some_path_with_dependencies/include
> 
> then I configure getfem with
> 
> ./configure --prefix=/some_installation_path/getfemseq BLAS_LIBS=-
> L/some_path_with_dependencies/lib -lopenblas -lpthread CPPFLAGS=-
> I/some_path_with_dependencies/include LDFLAGS=-
> L/some_path_with_dependencies/deps/lib LIBS="-lmumps_common -lpord -
> lmpiseq -lmetis -lesmumps -lscotch -lscotcherr -lpthread -lgfortran"
> --with-optimization=-O2 --with-pic --disable-matlab --enable-python -
> -enable-mumps
> 
> My "include" and "lib" dependencies folders contain:
> 
> image.pngimage.png
> This is the minimal amount of files that GetFEM links to. Depending
> on how you compile mumps, you might get slightly different
> dependencies, but I think you get the idea.
> 
> If you are still stuck after this, just let me know and I can tell
> you how to debug the configure error more systematically.
> 
> Best regards
> Kostas
> 
> On Mon, May 24, 2021 at 6:32 PM edgar <edgarlux@cryptolab.net> wrote:
> > Hi, Pierre. Disclaimer: I haven't been able to compile the test
> > suite 
> > correctly (take my advice with reserves).
> > 
> > The sequential libraries of MUMPS usually have _seq before the .o .
> > I 
> > can imagine that you already checked that you have the right
> > libraries. 
> > By the looks of it, you are using the parallel version, but I don't
> > dare 
> > to draw such conclusion. This comes from the configure file, and
> > you 
> > might want to try it:
> > 
> >      "-lsmumps_seq -ldmumps_seq -lcmumps_seq -lzmumps_seq"
> > 
> > I remember that I was able to hijack the library files with a
> > variable 
> > (MUMPS_LIBS?), but I am sure that the developers will have much
> > better 
> > advice than this.
> > 
> > Bonne chance!
> > 
> > On 2021-05-24 15:48, pierre.gosselet wrote:
> > > Dear all,
> > > I am working on a cluster and I need to compile getfem and most
> > of its
> > > dependency.
> > > 
> > > I have a hard time compiling the mumps support. I know close
> > questions
> > > were already asked but I did not find the answer. Sorry if I
> > missed it.
> > > 
> > > I compiled sequential mumps without trouble, examples work fine.
> > > 
> > > I tried to configure getfem with:
> > > ./configure --enable-mumps --with-mumps-include-
> > > dir="$MUMPS_DIR/include" --with-mumps="-fopenmp -lsmumps -
> > lmumps_common
> > > -ldmumps -lcmumps -lzmumps -lpord -lesmumps -lscotch -lscotcherr
> > -
> > > llapack -lmpiseq -lblas -lpthread" --with-pic
> > > 
> > > I always have the following problem:
> > > checking for library containing smumps_c... no
> > > configure: error: The function smumps_c couldn't be found in the
> > > provided MUMPS libraries.
> > > 
> > > The --with-mumps is inspired by the compilation line for mumps
> > examples
> > > (I have copied the libseq includes file in the include
> > directory).
> > > All the libraries are in the $LD_LIBRARY_PATH, but I also tried
> > with
> > > the full path. $MUMPS_DIR is well defined. I also tried with the
> > mumps
> > > module provided by the cluster's administrator (with adapted
> > include),
> > > same problem.
> > > 
> > > Can anyone help me ?
> > > 
> > > best regards
> > > pierre
> > 

-- 
Pierre Gosselet
CR CNRS (research agent), HdR
LaMcube, Univ.Lille/CNRS (UMR 9013)/Centrale Lille
Cité scientifique, Bat. Esprit, bureau S1.74
Avenue Paul Langevin, 59655 Villeneuve d'Ascq Cedex
Tel: +33 320434349