Installation error cp2k 8.2 at ELPA+GPU

445 views
Skip to first unread message

mandar.kul...@gmail.com

unread,
Jun 23, 2021, 4:37:25 PM6/23/21
to cp2k
Dear CP2K community,

I am installing cp2k 8.2 with toolchain options on Ubuntu 20.04 LTS with working CUDA drivers installed. (CUDA 11.2 + driver version 460.84)

./install_cp2k_toolchain.sh --gpu-ver=V100 --enable-cuda --with-openmpi=system --with-libxc=install --with-libint=install --with-fftw=install --with-openblas=install --with-elpa=install --with-libxsmm=install --with-plumed=install

During the installation of ELPA, I faced an error for the GPU version (../build/elpa-2020.11.001/build_gpu/config.log). I have attached the log file generated during elpa installation.

Any help or suggestions would be highly appreciated.

Best,
Mandar Kulkarni


elpa_config.log

Fabian Ducry

unread,
Jun 23, 2021, 4:51:17 PM6/23/21
to cp...@googlegroups.com

Hi Mandar

The error is "Could not link cublas; try to set the cuda-path or disable GPU support". This happens because configure is called without setting the path to cuda. To properly do that you can set the environment variable CUDA_PATH before running the toolchain, e.g. export CUDA_PATH=path_to_cuda. elpa then expects the cublas library in path_to_cuda/lib64

Cheers

Fabian

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/b9c7e7e1-e7b0-4785-a618-be21ee90c477n%40googlegroups.com.

Mandar Kulkarni

unread,
Jun 24, 2021, 12:33:52 PM6/24/21
to cp...@googlegroups.com
Hi Fabian,

Yes, CUDA_PATH was the issue, thanks for the suggestion.

But, still, I am facing problems at the "make" step. I added paths to NVCC and CUDA_PATH in local_cuda arch files.

then :

make -j 6 ARCH=local_cuda VERSION="sopt popt ssmp psmp" | tee make_cuda.log

I can install ssmp version, but facing an error for psmp executable.

This was a highlighted error:

/home/mandar/softwares/cp2k/cp2k-8.2/src/cp2k_info.F:18:7:

   18 |    USE input_constants,                 ONLY: id_development_version,&
      |       1
Fatal Error: Cannot open module file ‘input_constants.mod’ for reading at (1): No such file or directory
compilation terminated.

I have also attached the output of the make command. Please can you suggest what could be wrong?

Thanks in advance,
Mandar  Kulkarni


make_cuda.log
Reply all
Reply to author
Forward
0 new messages