Intel MPI and Compiler

[aroccon]

On Discoverer, there two known issues related to Intel components:

-Intel MPI for input and output does not seem to work (none of the MPI data representation work).

-Intel COMPILER (both IFX and IFORT) creates problematic executables.
The code seems to work fine only if the grid points/node ratio is kept relatively low.
On small number of nodes (1-4), the code crashes with segmentation fault.
The fault depends on the grid points/node ratio and is not a clear point on the code.

Possible reasons:
-Arrays pre-allocated (e.g. in. theta_check and helmoltz).

On the same machine, if the code is compiled with Nvidia, AMD or Gfortran no problems arise.

[aroccon]

Official response:
Intel MPI is known for not working with Fortran code compiled on AMD CPU with the Intel MPI compiler wrappers. Avoid that MPI library, unless your programming code is written C++ >=17. Sometimes compiling the code by using ifx instead of ifort via the MPI wrapper compilers produced working binary, but it is not a general rule.

Also note that Intel MPI is based on MPICH. Therefore, use MPICH in conjunction with Intel compilers and that will do the trick.