Charm++ parallel programming system

[Sam White <white67 AT illinois.edu>]

Hi Mert,

It looks like you are trying to build Charm++ on top of MPI. We generally prefer use of libfabric (OFI) as the most performant communication substrate to build Charm++ on Intel OmniPath network systems. To do that, build Charm++ with the following:

./build charm++ ofi-linux-x86_64 smp -j12 --with-production

If you do want to try an MPI build, then I think there is an inconsistency in your compiler modules and build command. It looks like you are trying to use Intel MPI with the GCC v10.2.0 compilers module loaded. I think the IMPI wrapper for GCC is properly named "mpicpc" on systems that support such configurations. So you can either try building with Intel MPI and GCC like this:

MPICXX=mpicpc ./build charm++ mpi-linux-x86_64 smp -j12 --with-production

Or with Intel MPI and the Intel compilers module loaded using:

MPICXX=mpiicpc ./build charm++ mpi-linux-x86_64 smp -j12 --with-production


Thanks, let us know if you run into any issues,
Sam

On Thu, Oct 28, 2021 at 2:14 PM mert gölcük <mertgolcuk AT gmail.com> wrote:

Dear developers,

I am trying to compile charm++ and NAMD on SUPERMUC-NG Supercomputers. NAMD is already installed on the Supermuc computers, but it is not working efficiently and fast when I compare it with the Stampede2. The architecture of both systems is the same. I attached the out file and input files. I couldn't solve the problem. Could you please help me with this issue?

Kind regards,

Mert Golcuk

System Specs;

CPU: 2x Intel Skylake Xeon Platinum 8174 

Network: Intel OmniPath

OS: SUSE Linux Enterprise Server 15 SP1