Dear users,
I am trying to install and build LAMMPS in Cray XC50.
I install lammps-stable.tar.gz, and unzip it.
To build LAMMPS, I take some commands.
############################################
cd lammps-stable/lib/gpu
vi Makefile.mpi
CUDA_HOME =/usr/local/cuda → CUDA_HOME =/usr/local/cuda-11.2
make -f Makefile.mpi
cd …/…/src
make yes-gpu
make mpi
#############################################
The above commands are on my desktop PC, not on xc50.
And, I can obtain lmp_mpi executable file.
It can operate on my desktop PC, but cannot operate on xc50.
When I perform it on XC 50, an error “libmpi_cxx.so.40: cannot open shared object file: No such file or directory” is displayed.
However, I cannot find libmpi_cxx.so.40 on both my desktop PC and xc50.
What is the difference between my pc and xc50?
And, how can I solve this problem?
Best regards
This is a question for the user support operating the Cray machine. The Cray Linux environment, depending on how the individual machine is set up, can have significant differences in how you should compile applications. LAMMPS assumes by default a conventional (desktop) installation or what is done on simple clusters. Particularly when using the GNU make based build system, manual edits of the relevant makefiles are expected and those need to be done based on the specific of a given system.
Please also consider using CMake to build LAMMPS. A properly installed CMake binary on your machine can simplify the process as it can auto-detect a lot of settings that are required to be done manually in the GNU make based build process. In that case you will need to download a newer patch release of LAMMPS since that adds the required support for CUDA-11, which was not (yet) released at the time of the last stable release of LAMMPS.
Thank you for your comments.
I try to build LAMMPS by CMake.
And, I’ll ask the support of Cray XC 50 to set LAMMPS.