[CP2K-user] [CP2K:22409] Re: CP2K 2026.2 on ubuntu 22.04 + oneAPI: The MPI_Comm_c2f() function was called before MPI_INIT was invoked.

[Martin Konôpka]

Dear Johann,

Thanks for your comment. Yes, it is so. And my yesterday's suspicion  that 
I omitted to source the
    /opt/intel/oneapi/2026.1/oneapi-vars.sh
script was not true; I did source the script. Later I even  repeated the  
whole toolchain + build procedure and the result was again a binary with 
links to multiple MPI libraries. And not only MPI. I can see also a 
non-oneAPI LAPACK and ScaLAPACK there. The mis-linked libraries are at 
least these:

   libscalapack-openmpi.so.2.1 => 
/lib/x86_64-linux-gnu/libscalapack-openmpi.so.2.1
   liblapack.so.3 => /lib/x86_64-linux-gnu/liblapack.so.3
   libmpi.so.40 => /lib/x86_64-linux-gnu/libmpi.so.40

I do now know where is a central config file in which I could disable 
linking the undesired libraries. It will perhaps be  simpler for me to 
uninstall the undesired MPI, LAPACK, etc libraries from the system. In the 
past I needed them because I was using a CP2K version  installed from a 
standard  ubuntu repository, similarly also a Quantum Espresso version. I 
will no more need the old versions from the ubuntu repository so that I can 
unistall them,  also the undesired libraries.

Best,
Martin.

Dátum: piatok 21. augusta 2026, čas: 16:06:53 UTC+2, odosielateľ: Johann 
Pototschnig

> The linker links to two mpi libraries, the oneapi one and 
> libmpi.so.40 => /lib/x86_64-linux-gnu/libmpi.so.40 (0x00007fa2f60c9000)
>
> This probably causes the problems. You need to remove the system mpi from 
> the linker, so it only links the oneapi one. 
>
>
> On Thursday, August 20, 2026 at 6:58:24 PM UTC+2 Martin Konôpka wrote:
>
>> Dear Frederick,
>>
>> Yes, you are right. Despite trying to be VERY careful, I perhaps (?) 
>> forgot to run the script
>>    /opt/intel/oneapi/2026.1/oneapi-vars.sh
>> in the shell in which I subsequently launched  the toolchain script etc. 
>> (Possibly I ran the intel script in a different shell mistakenly.) I now 
>> clearly see the problems when  I look at the output of the
>>     ldd /dalsie/cp2k-2026.2/install/bin/cp2k.psmp
>> commands.
>> I attach two ldd outputs, one with the correct build (Linux Mint Zena) , 
>> the  other with the incorrect build (Ubuntu 22.04).
>> I stress that I issued the ldd commands AFTER sourcing the scripts
>>      /opt/intel/oneapi/2026.1/oneapi-vars.sh
>>      /dalsie/cp2k-2026.2/install/cp2k_env
>>
>> I will rebuild the CP2K on  the ubuntu 22.04 and then I will do the 
>> regression tests. (I have  not tried them yet.) As for  Spack, I have  not 
>> used it for the CP2K 2026.2 build process. If possible to use,  I prefer 
>> the toolchain to Spack.
>>
>> Once more thank you very much for helping me.
>>
>> Best,
>> Martin.
>> Dátum: štvrtok 20. augusta 2026, čas: 17:26:13 UTC+2, odosielateľ: 
>> Frederick Stein
>>
>>> Dear Martin,
>>> I guess, CP2K picked up the system-provided OpenMPI-implementation in 
>>> some places (see the end of the cmake.log_Ubuntu_22.04.). Please 
>>> double-check with your Spack-log files. Have you tried to run the 
>>> regression tests beforehand? Without any further information it is hard to 
>>> tell which library picked up the wrong MPI-installation. You can add the 
>>> TRACE keyword to the &GLOBAL section of your input file and check where 
>>> CP2K fails. If you are lucky, you may even see where it fails because a 
>>> routine wasn't left.
>>> Best,
>>> Frederick
>>>
>>> Martin Konôpka schrieb am Donnerstag, 20. August 2026 um 16:54:24 UTC+2:
>>>
>>>> Dear Frederick,
>>>>
>>>> Thank you for the response. Yes, at the moment I have at hand all such  
>>>> files from my builds on the Linux Mint Zena and  on the Ubuntu 22.04. I 
>>>> attach them, also the log files from  cmake and the files (or only parts of 
>>>> them) which I created by copy-pasting my commands and  their standard 
>>>> outputs printed on the terminal.
>>>>
>>>> If needed, I can later send such  files form my build on Ubuntu 26.04 
>>>> (which is a notebook that I do not have at hand  at the  moment).
>>>>
>>>> BTW,  the toolchain script on 22.04 recommended to build the binaries 
>>>> with the  command
>>>>    ./build_cp2k.sh -j 18
>>>> In my actual build I used only 16 (not 18). I hope this is not the 
>>>> source of the reported problem.
>>>>
>>>> Best,
>>>> Martin.
>>>> Dátum: štvrtok 20. augusta 2026, čas: 15:56:02 UTC+2, odosielateľ: 
>>>> Frederick Stein
>>>>
>>>>> Dear Martin,
>>>>> Do you have log files from the build process? If you have a different 
>>>>> MPI implementation on your system, it could be that one of those were 
>>>>> picked up somewhere.
>>>>> Best,
>>>>> Frederick
>>>>>
>>>>> Martin Konôpka schrieb am Donnerstag, 20. August 2026 um 14:30:49 
>>>>> UTC+2:
>>>>>
>>>>>> Dear CP2K experts,
>>>>>>
>>>>>> I have  built CP2K 2026.2 on  several Linux workstations and I am  
>>>>>> trying to run it. In each case I used the Intel oneAPI 2026.1 compilers and 
>>>>>> libraries installed in the same way. Also the  CP2K was built on each of 
>>>>>> the workstations in the same way with one minor excpetion - see below. I 
>>>>>> used the toolchain. The results are different however:
>>>>>>
>>>>>> Ubuntu 26.04:  My CP2K build works.
>>>>>> Linux Mint Zena (based on ubuntu 24.04): My CP2K build works.
>>>>>> Ubuntu 22.04: My CP2K build fails with the message
>>>>>>
>>>>>> *** The MPI_Comm_c2f() function was called before MPI_INIT was 
>>>>>> invoked.
>>>>>> *** This is disallowed by the MPI standard.
>>>>>> *** Your MPI job will now abort.
>>>>>> [oslik:116792] Local abort before MPI_INIT completed completed 
>>>>>> successfully, but am not able to aggregate error messages, and not able to 
>>>>>> guarantee that all other processes were killed!
>>>>>>
>>>>>> There is no error message in the usual output file. I attach it, as  
>>>>>> well  as my script used to launch  the CP2K executable. As for the input 
>>>>>> file, it is the  same as in my recent post here:
>>>>>> https://groups.google.com/g/cp2k/c/_B6BE1Q78L0
>>>>>>
>>>>>> To configure the CP2K on the two  newer OSs, I issued the command
>>>>>>
>>>>>> ./install_cp2k_toolchain.sh  --mpi-mode=intelmpi  --math-mode=mkl 
>>>>>>  --libint-lmax=6  --with-ifx=yes  --with-cmake=system  --with-eigen=install 
>>>>>>  --with-dftd4=install
>>>>>>
>>>>>> On ubuntu 22.04 I used the  same with the exception of cmake; I used 
>>>>>> install instead of system, because the system cmake was too old.
>>>>>>
>>>>>> As for the oneAPI environment one the ubuntu 22.04, I successfully 
>>>>>> used it to build Quantum Espresso 7.5 (and  it works with the Intel MPI), 
>>>>>> so I really do not know what is the cause of the problem with the CP2K and would 
>>>>>> greatly appreciate your opinions and advices.
>>>>>>
>>>>>> Best regards
>>>>>> Martin Konôpka
>>>>>>
>>>>>

-- 
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+unsubscribe at googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/3a99fd01-a5a0-4cad-addd-8a0547944ef2n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20260821/7cf66bd0/attachment-0001.htm>