Error while compiling the code in SLURM based system

[Hnair13]

Hello,
I am a new user and I recently tried to install WarpX in the SCARF which is a SLURM based system. I could do the configuration but while compiling I got the following error which am not able to understand. Hope you can help me out here. I have checked the module dependencies and I do have all the required modules.

[Hnair13]

Hello,
I actually tried to compile again by loading all the modules including HDF5 and OpenMPI. If am not wrong the code was getting compiled but it threw an error. I am not sure what the error implies.

[RemiLehe]

Hi,
Thanks for your question.
It seems that some of the modules that you loaded are inconsistent. The error message seems to indicate that the MPI version used by the HDF5 module is different than the MPI version used in other modules.
Since every cluster is different and has their own specificities, would it be possible for you to contact your cluster administrator about this issue?

[ax3l]

I second @RemiLehe's assessment.

@Hnair13 can you please share the cluster details requested? Where are you computing and which software modules did you load? We can guide you if we have more details to find a consistent set :)

[Hnair13]

Hello,
I am using the SCARF cluster provided by STFC in the UK (url). I tried to compile the code using GCC 11.3.0 and OpenMPI 4.1.4.
The SCARF cluster consists of both CPU and GPU-based nodes. To configure code on the cluster, I have used the following modules:

1. GCC 11.3.0: The GNU Compiler Collection version 11.3.0.
2. OpenMPI 4.1.4-GCC 11.3.0: Open MPI version 4.1.4 compiled with GCC 11.3.0.
3. CMake/3.23.1-GCCcore 11.3.0: CMake version 3.23.1 compiled with GCC 11.3.0.
4. Git 2.36.0-GCCcore-11.3.0: Git version 2.36.0 compiled with GCC 11.3.0.

Additionally, I have found that the existing HDF5 version on the cluster is 1.12.0, which is compiled using GCC 11.3.0 and another MPI implementation, and I suspect that this might be causing issues.
I hope you could help me out here and do let me know if u need any other information.

[ax3l]

Thank you for the details @Hnair13!

Ok, so it looks like you currently compile for CPUs and the GPU nodes seem to use CUDA. Or are you trying to use the GPUs? :)

Yes, it sounds like the HDF5 version might be incompatible wit hthe other modules you listed, especially the GCC and MPI module, I suspect.

The best solution is to write a support ticket (email) to your cluster support, asking them for a compatible HDF5 module to the modules you listed already.

As work-around, load all modules but not the HDF5 module. You can compile HDF5 yourself like this:

# HDF5
git clone -b hdf5-1_14_1-2 https://github.com/HDFGroup/hdf5.git $HOME/src/hdf5
rm -rf $HOME/src/hdf5-build
cmake -S $HOME/src/hdf5 -B $HOME/src/hdf5-build -DBUILD_TESTING=OFF -DHDF5_ENABLE_PARALLEL=ON -DCMAKE_INSTALL_PREFIX=$HOME/sw/hdf5-1.14.1.2
cmake --build $HOME/src/hdf5-build --target install --parallel 8
rm -rf $HOME/src/hdf5-build

Instead of loading the HDF5 module, you then set this to find HDF5 during CMake configure and runtime:

# for compilation
export CMAKE_PREFIX_PATH=${HOME}/sw/hdf5-1.14.1.2:$CMAKE_PREFIX_PATH
# for running
export LD_LIBRARY_PATH=${HOME}/sw/hdf5-1.14.1.2/lib64:$LD_LIBRARY_PATH

Just to make communication and identification a bit easier, is it ok if you add your full name and affiliation to your GitHub profile? :) https://github.com/settings/profile