[CP2K-user] [CP2K:20765] Re: compilation problems - LHS and RHS of an assignment statement have incompatible types

[bartosz mazur]

Hi Frederic, 

I've used Intel OneAPI 2024.2. and it helped with the error we discussed. 
Thanks a lot for that! 

However, still some tests failed (correct: 4091 / 4227; failed: 136). Now 
most of the failed tests are killed without additional information with:

```
===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 0 PID 172367 RUNNING AT r23c03b11
=   KILLED BY SIGNAL: 11 (Segmentation fault)
===================================================================================

===================================================================================
=   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
=   RANK 1 PID 172368 RUNNING AT r23c03b11
=   KILLED BY SIGNAL: 11 (Segmentation fault)
===================================================================================
```

and sometimes also this message is printed:

```
LIBXSMM_VERSION: develop-1.17-3834 (25693946)
LIBXSMM_TARGET: clx [Intel(R) Xeon(R) Platinum 8268 CPU @ 2.90GHz]
Registry and code: 13 MB
Command (PID=172367): 
/lustre/pd01/hpc-kuchta-1716987452/software/cp2k/exe/local/cp2k.psmp 
2H2O_t01.inp
Uptime: 0.932725 s
```

or 

```
LIBXSMM_VERSION: develop-1.17-3834 (25693946)
CLX/DP      TRY    JIT    STA    COL
   0..13     22     22      0      0 
  14..23     15     15      0      0 
  24..64      0      0      0      0 
Registry and code: 13 MB + 320 KB (gemm=37)
Command (PID=132831): 
/lustre/pd01/hpc-kuchta-1716987452/software/cp2k/exe/local/cp2k.psmp 
admm_dbcsr_thread_dist.inp
Uptime: 1.272898 s
```

I was able to find similar issue (here 
<https://github.com/libxsmm/libxsmm/issues/805>) but I am not sure how I 
could fix it. I performed the regetests twice and in some cases the tasks 
finished without error the first time, but failed the second time, and the 
opposite: for example `QS/regtest-hfx/H2-ADMM-full.inp` was `OK` in the 
first run but finished with `RUNTIME FAIL` in second run, or 
`QS/regtest-as-1/h2_gapw_pp_2-4.inp` finished with `OK` in the first 
run (as the only one in this set) but in the second run finished with `RUNTIME 
FAIL`. In the attachment I provide outputs from toolchain, make and 
regtests 1st and 2nd run. 

The thing I've noticed is that toolchain is using ifort, which is some 
older version `ifort (IFORT) 2021.13.0 20240602`. Do you think using ifx 
would be better and maybe could help solving this issue? If yes, how can I 
force toolchain to use ifx instead of ifort? 

Another question - none of the regtests ended in `WRONG`. Does this mean 
that I can assume that cp2k is safe to use and if an error occurs, the job 
will be killed instead of getting an erroneous result? 

Best
Bartosz

wtorek, 8 października 2024 o 15:46:14 UTC+2 Frederick Stein napisał(a):

> Hi Bartosz,
> No, Intel 2021 will be probably not work, it is older than Intel 2022. I 
> meant something like Intel OneAPI 2023 or 2024.
> Best,
> Frederick
>
> bartosz mazur schrieb am Dienstag, 8. Oktober 2024 um 14:43:19 UTC+2:
>
>> Hi Frederick, 
>>
>> Thank you for your quick response! Just to be sure, if I compile the 
>> latest version of cp2k using Intel 2021 (
>> https://www.cp2k.org/dev:compiler_support), I should no longer have the 
>> problems described? I ask because I don't see a module with Intel OneAPI 
>> 2024 on our HPC, so I am considering using either an older module or asking 
>> the admins to provide a newer one.
>>  
>> Best
>> Bartosz
>>
>> wtorek, 8 października 2024 o 14:07:15 UTC+2 Frederick Stein napisał(a):
>>
>>> Dear Bartosz,
>>> If you want to compile with Intel, then drop the "--with-gcc" flag. 
>>> Regarding Intel, we do not test Intel 2022.2 anymore. You should try the 
>>> IntelOneAPI containing more recent compilers instead.  We are currently 
>>> testing version 2024.2.
>>> The warnings can be ignored for now, but we are aware of that issue and 
>>> will make adjustments later after dropping some older compilers.
>>> Regarding the runtime errors. The error "LHS and RHS of an assignment 
>>> statement have incompatible types" could be a compiler bug (see 
>>> https://community.intel.com/t5/Intel-Fortran-Compiler/Segmentation-fault-due-to-assignment-of-derived-type-variable/td-p/1489823). 
>>> The allocation error may also be a compiler bug as the respective array is 
>>> always allocated and the routine is left directly after deallocating the 
>>> array earlier in the routine.
>>> Best,
>>> Frederick
>>>
>>> bartosz mazur schrieb am Dienstag, 8. Oktober 2024 um 13:17:08 UTC+2:
>>>
>>>> Hi all, 
>>>>
>>>> I recently managed to compile cp2k on our cluster, but regtests showed 
>>>> several errors. Most of the failures are due to the error `forrtl: 
>>>> severe (189): LHS and RHS of an assignment statement have incompatible 
>>>> types` or `forrtl: severe (153): allocatable array or pointer is not 
>>>> allocated`. After looking at the output from `make` I noticed that 
>>>> there are quite a few similar warnings there:
>>>>
>>>> ```
>>>> /lustre/pd01/hpc-kuchta-1716987452/software/cp2k/exts/dbcsr/src/mpi/dbcsr_mpiwrap.F(1930): 
>>>> warning #8100: The actual argument is an array section or assumed-shape 
>>>> array, corresponding dummy argument that has either the VOLATILE or 
>>>> ASYNCHRONOUS attribute shall be an assumed-shape array.   [MSGIN]
>>>>          CALL mpi_isend(msgin, msglen, MPI_LOGICAL, dest, my_tag, &
>>>> ------------------------^
>>>> ```
>>>>
>>>> For compilation I used GCC 12.2.0 and intel 2022.2.1. My toolchain 
>>>> command was `./install_cp2k_toolchain.sh --mpi-mode=intelmpi 
>>>> --with-intel --with-gcc=system --with-plumed --with-quip --with-pexsi 
>>>> --with-ptscotch --with-superlu --with-fftw=no --with-hdf5`. In the 
>>>> attachment I provide all outputs from toolchain, make, and regtests. 
>>>>
>>>> I'm not sure what went wrong and how should I proceed so any help will 
>>>> be much appreciated! 
>>>>
>>>> Best
>>>> Bartosz
>>>>
>>>

-- 
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 on the web visit https://groups.google.com/d/msgid/cp2k/00fddcc1-3a8d-41fa-a5bf-d30c5b637b45n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20241011/1fd37ee9/attachment-0001.htm>