Problems with parallel Dalton and DFT

[luca]

Dear Dalton,
I receive an error whenever I try to run any DFT calculation in parallel. Let me give you the details of my installation.

Dalton 2013.2
Intel compilers + MKL vers. 13.0.1.117
Open MPI vers. 1.6.5
cmake vers 2.8.9

Dalton was compiled using:
./setup --mpi --mkl=parallel --int64 --fc=mpif90 --cc=mpicc --cxx=mpicxx --type=debug

I used type=debug to try and obtain further info on the error.

I use the following input files:

test.dal

**DALTON INPUT
.RUN WAVE
**WAVE FUNCTIONS
.DFT
B3LYP
**END OF DALTON INPUT

test.mol

BASIS
cc-pVDZ
this is a comment line
this is a comment line
1
10. 1
Ne 0.0 0.0 0.0

If I run this calculation in a serial environment (that is, using the code compiled with MPI support but on 1 cpu on 1 node), it successfully completes.
If I try to run this calculation on 2 cpus I get the following not so informative error:

Error in /users/software/kemi/openmpi-1.6.5/bin/mpiexec -mca plm_rsh_disable_llspawn 1 -np 2 /users/software/kemi/Luca/DALTON-2013.2-Source-parallel/build/dalton.x, exit code 15

and the following error from the queue:

[node036.xxx:21850] *** An error occurred in MPI_Recv
[node036.xxx:21850] *** on communicator MPI_COMM_WORLD
[node036.xxx:21850] *** MPI_ERR_TRUNCATE: message truncated
[node036.xxx:21850] *** MPI_ERRORS_ARE_FATAL: your MPI job will now abort

The Dalton output stops at:

Automatic occupation of symmetries with 10 electrons.

Iter Total energy Error norm Delta(E) SCF occupation
-----------------------------------------------------------------------------
1 Screening settings (-IFTHRS, DIFDEN) -5 F

without any further message. It appears to me that there is some problem of I/O connected with OpenMPI, but I'm not able to understand exactly what. Moreover, it seems to be strongly related to DFT, since I tried many other tests from the test suite and only those involving DFT failed.

Any idea?

Thanks for any help, have a nice day
/Luca

[bast]

hi Luca,
this looks to me like an integer type mismatch between the C code
(DFT) and Fortran (rest of Dalton). unfortunately i cannot investigate this further
at this stage but i will file a ticket in the bugtracking to verify it.
the easy workaround is to not use 64bit integers, do you need them?
best regards,
radovan

[luca]

Hi Radovan,
Unfortunately yes, we would like to have 64bit integers. Is there any keyword I can add to setup in order to have a more complete error output or any way to increase the print level?
Thanks!
/Luca

[luca]

Hi Radovan,
I tried to compile Dalton in parallel but without --int64. Unfortunately I receive the same error as before. However, I did not change the OpenMPI libraries, which are still compiled with 64bit integers.
/Luca

[bast]

hi Luca,
for 32bit integer Dalton you should use 32bit integer OpenMPI.
unfortunately i am a bit busy right now but i will try to reproduce the problem with 64bit integers
next week. you have sent good information to reproduce it but just cannot do it right now.
best wishes,
radovan

[luca]

No problem and no hurry. I'll try and compile also a 32bit version of openmpi.
Best!
/Luca

[juanjoaucar]

Hi everyone,

I'd like to know if this issue was solved out.
I have the same problem: I get the error described on this thread whenever I want to do parallel calculations using DFT, with DALTON compiled on 64bits.
I tried compiling the Dalton Code with omp (as suggested here) but the error persists. By the way, I'm using the 2018 version.

Thanks in advance,
Juan

[magnus]

Unfortunately there are still issues with 64-bit integers. However, I think they a rarely needed for DFT calculations in Dalton. What kind of issue are you having with a 32-bit integer compilation?

[juanjoaucar]

Unfortunately there are still issues with 64-bit integers. However, I think they a rarely needed for DFT calculations in Dalton. What kind of issue are you having with a 32-bit integer compilation?

Thanks for the reply. I don't really need those kind of calculations. I was just doing some tests on my 64-bit integer compiled version, and when I've found this error I thought I did something wrong during the installation process.
I assume now that it's something not solved for the 64-bit integers compilation. Thanks!