My ifort classic built of v2024.1 gives segfaults with TC-LRC + ADMM.

[Martin Konôpka]

Dear CP2K experts,

I compiled CP2K 2024.1 with Intel OneAPI tools, using the CP2K toolchain. (I acknowledge that the toolchain is now quite easily usable compared to those some years ago. I prefer to use the toolchain for it provides flexibility.)  Resulting executable cp2k.popt passed successfully a test with the PBE functional. However, it crashed upon a segmentation fault when I used PBE0 with the TC-LRC and ADMM techniques. The same input file (an oxygen atom) was processed cleanly on the same machine using CP2K 9.1 installed through the OS package manager . For each of the tests I cautiously set the environments so that v9.1 used the "factory" tools (installed with the ubuntu apt tool) and v2024.1 used the Intel tools from /opt/intel/oneapi.

I would be grateful for your advices on what should I do in a different way.

Some additional information:

OS: lubuntu 22.04, quite a standard installation

256 GB RAM in the machine

I attach:

* a file with my command to the toolchain

* the arch file local.psmp

* output from the command ldd /dalsie/cp2k-2024.1/exe/local/cp2k.popt

* cp2k.inp

* cp2k.out_v9.1

* cp2k.out_v2024.1

* stderr_v2024.1

If more info is needed from my side, just let me know. Thanks.

Best regards

Martin Konôpka.

[Krack Matthias]

Hi Martin

 

It seems that you are using the default OMP_STACKSIZE. I suggest to increase that value to at least 16M or more, e.g. by using “export OMP_STACKSIZE=64M”.

CP2K v2024.1 uses much more OpenMP constructs than v9.1 which causes that need for a larger OMP_STACKSIZE.

 

HTH

 

Matthias

 

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/d04cd014-76bd-4a28-903a-6ea76b5ab085n%40googlegroups.com.

[Martin Konôpka]

Hi Matthias,

Thanks. Unfortunately, the job crashes quite in the same way also after I have set the OMP_STACKSIZE variable as you suggested. I think I will try to use the non-classic intel compilers (although I have no idea about differences) and perhaps will omit some libraries from the built.

Best

Martin.

Dátum: utorok 9. júla 2024, čas: 16:41:40 UTC+2, odosielateľ: Krack Matthias

[Krack Matthias]

Hi Martin

 

The new Intel Fortran compiler ifx (instead of the “classic” ifort) is not yet working with CP2K (see this issue, icx/ifort works but not icx/ifx). We hope that an upcoming oneAPI release (after 2024.2.0) will resolve the ICE.

I still suspect some kind of memory problem (“ulimit -s unlimited” is set?) in your case.

 

Best

 

Matthias

 

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/b87b83d9-857a-4425-8be1-1a2e79468a04n%40googlegroups.com.

[Martin Konôpka]

Dear Matthias,

You are right! The “ulimit -s unlimited” command did it!

First I did not use it because the man 2 ulimit page says

“Unimplemented system calls”

and I got confused. Still, it helped now and the speed of the exectutable is about the same as of the CP2K 9.1.

Great thanks!

Martin

Dátum: utorok 9. júla 2024, čas: 17:35:03 UTC+2, odosielateľ: Krack Matthias