Seg. fault. on yambo-4.2.1

[robertap]

Dear Daniele,
thanks a lot for taking the time to run the this calculation.
Both gvector.dat and gkvectors.dat are written, and apparenty correctly (they end with the same line </G-VECTORS> and </GK-VECTORS> as in the calculations that runs without problem on another machine).

On the other hand the fortran error suggests it might be a format issue, right? Since it is [ERROR]Error in qexml_read_planewaves fmt error ierr: 2

In any case I'll keep looking into it.
Thanks again
Roberta

[robertap]

Sorry again,

now I have another segmentation fault in the calculation of RPA epsilon of hBN. You helped solve the problem before by using --enable-int-linalg, and the calculation ran smoothly for the exchange self operator, but now it fails again at the RPA level.

<---> [01] CPU structure, Files & I/O Directories
<---> CPU-Threads:1(CPU)-12(threads)
<---> [02] CORE Variables Setup
<---> [02.01] Unit cells
<---> [02.02] Symmetries
<---> [02.03] RL shells
<---> [02.04] K-grid lattice
<---> [02.05] Energies [ev] & Occupations
<---> [03] Transferred momenta grid
<01s> [04] Optics
<01s> [LA] SERIAL linear algebra
<01s> [DIP] Checking dipoles header
<01s> [WF] Performing Wave-Functions I/O from ./SAVE
<01s> [FFT-X] Mesh size: 12 12 27
<01s> [X-CG] R(p) Tot o/o(of R) : 124 1152 100
<01s> Xo@q[1] | | [000%] --(E) --(X)[froggy2:08294] *** Process received signal ***
Segmentation fault (core dumped)


At this point I have yambo-4.2.1 installed in two machines (both use qe-6.1 outpus), in one machine it runs fine in serial and parallel (just tested for 2 cores) for all examples of hBN, but fails to convert the qe-6.1 output of a system I an interested in. In another machine, it converts correctly this system with p2y but the yambo calculations have systematically seg. fault errors (already at the hBN setp, epsilon).

At this point my wish is to have it work properly in at least one machine

Maybe I can share the config.log of the installation that gives seg. fault error? It was compiled using internal blas, and external netcdf and fft.

I have also compiled it using exclusively internal libraries. And same issue.

I would appreciate any advice, tests to make, things to look, recompilation ect...

Thanks in advance
Roberta

[Daniele Varsano]

Dear Roberta,
this sounds really strange, can you please post your input and report file?
what happens with a simple
./configure --enable-int-linalg
(a make clean_all is needed before recompiling).

Daniele

[robertap]

dear Daniele,

the seg. fault seems to be due to OpenMP.

Using intelmpi/14, ifort(14) and open-mp, the compilation is OK but execution gives seg. fault at various levels depending on the system (BN or larger system). By removing the open-mp option and keeping everything else the same AT COMPILATION everything works. Even if open-mp is not used for the calculation but still enabled during compilation, the seg. fault error still occurs.

Thanks
Roberta

[claudio]

Dear Roberta

I got a similar error with ifort13 + openmpi,
it seems there is a problem in the present open-mp implementation of yambo
with some systems.

the only solution is to compile without open-mp

Best
Claudio

[Davide Sangalli]

Dear Roberta,
I'll look into this problem of the compilation with OpenMP.

Did it work compiling yambo without OpenMP as Claudio suggested ?

At this point I have yambo-4.2.1 installed in two machines (both use qe-6.1 outpus), in one machine it runs fine in serial and parallel (just tested for 2 cores) for all examples of hBN, but fails to convert the qe-6.1 output of a system I an interested in. In another machine, it converts correctly this system with p2y but the yambo calculations have systematically seg. fault errors (already at the hBN setp, epsilon).

At this point my wish is to have it work properly in at least one machine

Meanwhile if you are able to convert on one machine and run on another, another solution could be to convert the DBs and then move on the machine where yambo is working.
The NETCDF DBs are not machine dependent.

For the conversion error I suspect it is due on how pwscf is compiled on the super-computer.
I think it may not write some info needed by yambo in same cases and then the conversion does not work properly.

Best,
D.

[robertap]

Dear Davide,

thank you for your feedback.

Yes, after removing the openmp option at compilation everything works fine. Although I have very big systems I do have access to nodes with large memory so I guess that is not a problem at the moment (I haven't done BSE yet, only GW).

Regarding the other compilation where conversion fails, I will try to copy the conversion as you suggest (or the pwscf output files and convert them). I guess once I'll be back from the conference I will be able to see if yambo-4.2.1 works when compiled with openmp and a more recent intel version (intel/17 is available on this machine).

thanks again for your help

Roberta