LSDalton compilation problem

[ehorkel]

Dear forum members,

as a board "Installing LSDalton" does not exist, I hope this board is the most appropriate for my question.

System Information: Rocks cluster 6.2, which is based on CentOS 6.x; the HPC Roll is installed, openmpi (1.6.2) and infiniband working fine.

uname -a gives:
Linux itchy.ias.tuwien.ac.at 2.6.32-504.16.2.el6.x86_64 #1 SMP Wed Apr 22 06:48:29 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux

I have already compiled and installed Dalton successfully using the following commands:

./setup --mpi --prefix=/share/apps
cd build/
make -j4
ctest -j4
make install

I tried to compile LSDalton using the same procedure, however, after starting "make -j4" the compilation starts but stops with several errors; the executables are not generated.

I have attached the compilation output in a PDF, I hope this is fine. Any tips/suggestions are warmly welcome.

Many thanks in advance,

best regards

Ernst

[tkjaer]

I only seee one error associated with quartic precision lapack, which I do not understand

I suggest you export the MATH_ROOT variable

export MATH_ROOT=$MKLROOT

if you have MKL installed (you can get a non-commercial version of MKL for free) or set it to some other math library path.

this is anyway much faster then using the internal lapack library provided by LSDalton
then perform the setup command again

[taylor]

I second Thomas's suggestion. I replied to a posting ages ago where someone had a problem and there was some request for a quad precision library (something that is not only not needed but highly undesirable from a performance point of view!). This general issue was discussed in a (non-Dalton context) posting on some website I had found but I can't find that link at the moment, sorry. Nevertheless any build that claims to need quad precision for anything has either an environment or setup problem, and starting by ensuring you have the right MATH_ROOT is an essential step.

Best regards
Pete
P.S. As Thomas implied, the MKL library is gender-neutral with respect to compilers: it can be linked against from gfortran/gcc as well as from ifort. But OpenMPI is not. It must be compiled with the compiler you plan to ultimately use for your application code.

[ehorkel]

Hello Thomas and Taylor,

thanks for the quick replay, exporting MATH_ROOT did the job.

Thanks,

Ernst