Compilation issues

Dear MRCC administrators,

I am writing to seek suggestions on resolving a compilation issue that I have come across. I am trying to compile MRCC with ifort version 17, with the Intel MPI library. To be more specific, here are the options that I chose while running build.mrcc:

./build.mrcc Intel -i64 -pOMP -pMPI=IntelMPI

I see the following error messages:
Space exceeded in Data Dependence Test in hgi_int_out_
Subdivide routine into smaller ones to avoid optimization loss

and

Space exceeded in Data Dependence Test in hgi_int_out_rangesep_
Subdivide routine into smaller ones to avoid optimization loss

Also, I would like to compile MRCC with the GNU compiler (gcc version 8.3.0, for example). But this mode of compilation is met with several errors with the options mentioned above.

I would greatly appreciate suggestions from your side.

Thanks in advance,
Dipayan Datta

Hi Dipayan,
I frequently ran into space issues when compiling MRCC, until I started compiling on a development node (a node where I can actually run serious calculations, not just submit jobs). The space and time requirements to compile MRCC sometimes necessitate having access to a fair amount of resources.

I'm not sure how to solve your problem with the GNU compiler though.

By the way it's cool that your question is marked as issue #1000. A new milestone for the MRCC forum!

With best wishes,
Nike

Hello Nike,

I am trying to compile MRCC on a national supercomputing facility. Thus, by no means I have access to a compute node. This is a good suggestion though and I will try to compile on a compute node of a local computer cluster, given the slowness of the compilation process.

Thanks,
Dipayan

Hi Dipayan,
Maybe that's a good idea: You can test it on your local cluster and see if it works. Then you can copy the binary files over to the national HPC centre, or you can at least see how many resource requirements you needed for compiling it on the local cluster.

On a national HPC centre, there might be a few development nodes that are available for testing (but they might be limited to maybe 30 minutes, which I think will not be enough).

You could submit a job to the queue, which will give you access to a compute node, then compile it there. 1 day and 32GB of RAM will be much more than enough.

Another thing you can do is submit an "interactive" job to the queue. On SLURM this can be done with the srun command: hpc-uit.readthedocs.io/en/latest/jobs/interactive.html and on PBS/TORQUE it can be done with qsub -I.

With best wishes!
Nike