[CP2K-user] issues installing cp2k v.7.1 on a Beowulf cluster with CentOS 7.6
Katharina Doblhoff-Dier
kdob... at gmail.com
Thu Feb 13 18:05:01 UTC 2020
Dear Sergey,
Thank you for your reply!
First of all: Let me note that the strange timing error that I got did not
appear when I did the same thing again. I do not know what happened there.
But yes, indeed i did get stuck on libint!
Unfortunately, we really do not have a newer compiler on our cluster. I
will try to install one.
Thanks again,
Best regards,
Katharina
On Wednesday, February 12, 2020 at 6:22:18 PM UTC+1, Sergey Chulkov wrote:
>
> Hi Katharina,
>
> You can compile CP2K 7.1 on CentOS 7.6 with the system-default compiler
> (gcc 4.8.5) using the following toolchain options:
>
> ./install_cp2k_toolchain.sh -j 16 --with-gcc=system --with-cmake=install
> --with-openmpi=system --with-libxc=install --with-libint=no
> --with-fftw=install --with-openblas=install --with-scalapack=install
> --with-libxsmm=install --with-elpa=no --with-sirius=no --with-gsl=install
> --with-libvdwxc=no --with-spglib=install --with-hdf5=no --with-spfft=no
>
> Please notice that gcc 4.8.5 has a limited support of C++11 standard,
> hence you will not be able to compile some vital libraries (e.g. libint2).
> If gcc 4.8.5 is the only compiler available on your cluster (try to run
> 'module avail' first), you can still build a recent gnu compiler manually.
> In case of a CentOS 7.6 cluster with Infiniband interconnect, gcc 9.2.0
> along with openucx 1.6.1 and openmpi 4.0.2 works without any issues. In
> order to compile openucx with Infiniband support, the package
> 'rdma-core-devel' from the CentOS 'base' repository needs to be installed.
>
> Best wishes,
> Sergey
>
> On Wednesday, February 12, 2020 at 9:35:47 AM UTC, Katharina Doblhoff-Dier
> wrote:
>>
>> Dear CP2K community,
>> I am having some issues getting cp2k to work on a Beowulf cluster running
>> CentOS 7.6 using gcc 4.8.5
>>
>> The first issue that I ran into was that toolchain had problems with
>> spfft. Looking into that, it seems spfft can only be compiled with gcc >=6
>> (a warining instead of a weird error message for the toolchain would be
>> nice!). Since I am not in dire need of plane waves, I kicked sirius and
>> spfft out of the toolchain via with-sirius=no and with-spfft=no. I did the
>> smae with libint and elpa, because I had issues with them too and
>> proceeded, just to get another rather weird error message from toolchain:
>>
>>
>> >./install_cp2k_toolchain.sh --with-openmpi=system --with-sirius=no --with-spfft=no --with-libint=no --with-elpa=no
>> MPI is detected and it appears to be OpenMPI
>> nvcc not found, disabling CUDA by default
>> Compiling with 64 processes.
>> ==================== Finding GCC from system paths ====================
>> path to gcc is /usr/bin/gcc
>> path to g++ is /usr/bin/g++
>> path to gfortran is /usr/bin/gfortran
>> Found include directory /usr/include
>> Found lib directory /usr/lib64
>> Step gcc took 0.00 seconds.
>> ==================== Getting proc arch info using OpenBLAS tools ====================
>> OpenBLAS-0.3.6.tar.gz: OK
>> Checksum of OpenBLAS-0.3.6.tar.gz Ok
>> OpenBLAS detected LIBCORE = zen
>> OpenBLAS detected ARCH = x86_64
>> ==================== Installing CMake ====================
>> cmake-3.15.1.tar.gz: OK
>> Checksum of cmake-3.15.1.tar.gz Ok
>> Installing from scratch into /home/mark/CP2K/cp2k-7.1/tools/toolchain/install/cmake-3.15.1
>> Step cmake took 231.00 seconds.
>> Step valgrind took 0.00 seconds.
>> ==================== Finding OpenMPI from system paths ====================
>> path to mpirun is /usr/local/openmpi/bin/mpirun
>> path to mpicc is /usr/local/openmpi/bin/mpicc
>> path to mpif90 is /usr/local/openmpi/bin/mpif90
>> path to mpic++ is /usr/local/openmpi/bin/mpic++
>> Step openmpi took 0.00 seconds.
>> Step reflapack took 0.00 seconds.
>> ==================== Installing OpenBLAS ====================
>> OpenBLAS-0.3.6.tar.gz is found
>> Installing from scratch into /home/mark/CP2K/cp2k-7.1/tools/toolchain/install/openblas-0.3.6
>> patching file kernel/x86_64/KERNEL.SKYLAKEX
>> Step openblas took 86.00 seconds.
>> ==================== Installing FFTW ====================
>> fftw-3.3.8.tar.gz: OK
>> Checksum of fftw-3.3.8.tar.gz Ok
>> Installing from scratch into /home/mark/CP2K/cp2k-7.1/tools/toolchain/install/fftw-3.3.8
>> Step fftw took 49.00 seconds.
>> Step spfft took 0.00 seconds.
>> Step libint took 0.00 seconds.
>> ==================== Installing LIBXC ====================
>> libxc-4.3.4.tar.gz: OK
>> Checksum of libxc-4.3.4.tar.gz Ok
>> Installing from scratch into /home/mark/CP2K/cp2k-7.1/tools/toolchain/install/libxc-4.3.4
>> Step libxc took 244.00 seconds.
>> Step libsmm took 0.00 seconds.
>> ==================== Installing Libxsmm ====================
>> libxsmm-1.14.tar.gz: OK
>> Checksum of libxsmm-1.14.tar.gz Ok
>> Installing from scratch into /home/mark/CP2K/cp2k-7.1/tools/toolchain/install/libxsmm-1.14
>> Step libxsmm took 13.00 seconds.
>> ==================== Installing ScaLAPACK ====================
>> scalapack-2.0.2.tgz: OK
>> Checksum of scalapack-2.0.2.tgz Ok
>> Installing from scratch into /home/mark/CP2K/cp2k-7.1/tools/toolchain/install/scalapack-2.0.2
>> Step scalapack took 35.00 seconds.
>> Step elpa took 0.00 seconds.
>> Step ptscotch took 0.00 seconds.
>> Step parmetis took 0.00 seconds.
>> /home/mark/CP2K/cp2k-7.1/tools/toolchain/scripts/tool_kit.sh: line 12: /usr/bin/date: Argument list too long <<<<<====== ?!?
>>
>> The problem seems to be a call to time, but this is weird, because the command itself works and has obviously been used before. So now I am confused. Any help is appreciated!
>> Thank you and best regards,
>> Katharina
>>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20200213/d1a41ff1/attachment.htm>
More information about the CP2K-user
mailing list