Installation advice

[Sebastian Muraru]

Hi, I'm setting up a workstation Fedora 34, cuda-11.4 (GTX 1060 6GB) , gcc 9.4.

I installed openmpi (4.1.0) and openmpi-devel through dnf. I use module load mpi to load it.

The toolchain script claims it does not detect it and installed openmpi-4.0.5 .

Should I expect any issue with the generated arch file in regards to openmpi?

Also, at the end, it claims nvcc not detected. Basically, I run this and get as below:

sudo ./install_cp2k_toolchain.sh --with-openmpi --enable-cuda --gpu-ver=P100

WARNING: (./install_cp2k_toolchain.sh, line 330) No MPI installation detected on your system. Ignore this message if you are using Cray Linux Environment
nvcc not found, disabling CUDA by default
Compiling with 40 processes.

...

==================== generating arch files ====================
arch files can be found in the /home/smuraru/Installed/cp2k-support-v8.2/tools/toolchain/install/arch subdirectory
ERROR: (./scripts/generate_arch_files.sh) cannot find nvcc, please check if cuda is installed or in system search path
ERROR: (./scripts/generate_arch_files.sh, line 114) Non-zero exit code detected.

Can someone aid me slightly?

[Alin Marin Elena]

it seems you miss both mpif90 and nvcc from your paths.

mpif90 --show
which mpif90
which nvcc
same terminal you run the installer script

personally I will not use sudo to install cp2k there is no need

I do not remember if fedora uses modules by default to manage the mpi
installations.

Without Questions there are no Answers!
______________________________________________________________________
Dr. Alin Marin ELENA
http://alin.elena.space/
______________________________________________________________________

> --
> 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/afd80ebc-5eb6-4fd8-86b2-3c1034e90628n%40googlegroups.com.

[Sebastian Muraru]

Thank you (multumesc mult!) for the answer!

For some reason, my MPI and CUDA were not detected. I installed cuda at a non-standard location and I loaded MPI through module load mpi.

Given you said "I will not use sudo to install cp2k there is no need" I figure I'd try installing from root and, magically, both MPI and nvcc were found.

During the "make -j 40..." phase my cuda.h, cuda_runtime.h, cublas_v2.h were not found - I think due to my non-standard location for the cuda installation, so I updated SYS_INCLUDE_PATH and SYS_LIB_PATH in common_vars.sh (in tools/toolchain/scripts) with the corresponding CUDA folders.