Fix for Intel Systems in CS8.3 (Git Commit 7342705)

[StandardRANSUser001]

Dear Yvon,

I just wanted to verify something that I have observed on my version of CS8.3 on an Intel system, after updating my code to commit 7342705 (Main branch of git.).

The solver seems to be hanging on the last timestep, and I wanted to see whether you observed this on your system as well?
I haven't run any cases for a few weeks but I don't remember observing this before updating this change. Have you observed this as well?

I am trying to avoid consuming excess HPC resources, so I would like the solver to "finish" rather than hang on the final timestep. I am not sure if this is relevent, but this was observed on a machine with pbs submission scripts.

As always, any assistance with this would be greatly appreciated.

Best regards,
Sean Hanrahan

[Yvan Fournier]

Hello,

No, to my knowledge, we have not observed this on our machines. Which version of the Intel compilers do you use (on our side, We mainly use the 2022.2 LLVM-based compilers on one cluster, and the (much) older 18.1 on another machine.
I also tested with 24.1 on my local workstation.

The PBS aspect might be relevant (we have in the past observed some strange side-efffects such as a memory leak with some versions of SLURM, so although such side effects may be rare, they are not unheard of).

Best regards,

Yvan

[StandardRANSUser001]

Hi Yvon,

So the code seems to hang after "Destroying structures and ending computation" in the run_solver.log files.

The problem is that the code keeps running in PBS, so I am being billed, even through the code is no longer running.

Local case configuration:

Date: Mon Apr 7 15:52:37 2025
System: Linux 4.18.0 Rocky Linux 8.10 (Green Obsidian)
Machine: XXXXXXX
Processor: model name : Intel(R) Xeon(R) Platinum 8470Q
Memory: 515648 MB
User: XXXXXXXXX
Directory: XXXXXXXXX
MPI ranks: 416 (appnum attribute: 0)
MPI ranks per node: 104
OpenMP threads: 1
Processors/node: 1

Compilers used for build:
C compiler: Intel(R) oneAPI DPC++/C++ Compiler 2022.0.0 (2022.0.0.20211123)
C++ compiler: Intel(R) oneAPI DPC++/C++ Compiler 2022.0.0 (2022.0.0.20211123)
Fortran compiler: ifx (IFORT) 2022.0.0 20211123

MPI version: 3.1 (Open MPI 5.0.5)
OpenMP version: 5.0

External libraries:
PT-SCOTCH 7.0.4

I/O read method: collective MPI-IO (explicit offsets)
I/O write method: collective MPI-IO (explicit offsets)
I/O rank step: 1

[Yvan Fournier]

Hello,

In this case, I would recommend either of 2 options :

- Apply a revert of that patch on the sources for your build
- Or use the gfortran Fortran compiler instead of the Intel compiler (that should be compatible with the Intel C/C++ compilers). Since very little Fortran code is left, performance should be the same.

Best regards,

Yvan