VASP6.3 fails for ncl executable (segmentation fault/error from radial.f90)

[sophie_weber]

Hello,

I have successfully run several VASP6.3 calculations using the collinear, std executable. However, when trying to run with the system including spin-orbit coupling and the non collinear ncl executable, I get an error having to do with a segmentation fault. I contact the Helpdesk for the supercomputer I am using and he responded with this:
-----------------------
According to the backtrace information reported in the core file dumped when the job crashed, the program stopped in radial::rad_gga_asph:

Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x00000000006c4d62 in radial::rad_gga_asph () at radial.f90:1318
The line referenced in the backtrace usses a loop to store the potential, therefore it seems that the segmentation fault reported is actually occurring at this point:

! multiply potential with Y_00 and store it
POTXC(1:R%NMAX,1,:)=POTXC(1:R%NMAX,1,:)*(1.0_q/SCALE)
IF (LPOT) THEN
DO LM=1,LMMAX_CALC
DO K=1,R%NMAX
! POTXC must be multiplied by Y_00 to get V_xc(r)
POT(K,LM,:)=POT(K,LM,:)+YLM(NP,LM)*SIM_FAKT*POTXC(K,1,:)
ENDDO
ENDDO
ENDIF
I could only find this thread on the VASP user forum referencing the source file radial.F, however the user reported that the error occurred running vasp_std in that case, therefore it might be unrelated.

May I suggest to report your specific case in the VASP user forum? The description above might help other users and hopefully the developers may suggest a workaround.
--------------------
For what it's worth, someone else in my group tried running the VASP6.3 executable as well on a completely different system, and it failed with the same error, so it does not seem to be system or calculation dependent.

If someone could suggest how to fix this so that it's possible to run ncl calculations, I'd greatly appreciate it. Let me know if you need any other files from me.

[fabien_tran1]

Hi,

Could you please provide the files of a case (the smallest) that is failing and specify the compiler.

[sophie_weber]

Hello,

Sorry, I didn't see this before; thank you much for your help. I attach the files, as well as the job submit script for the supercomputer, in the tar file.
This is the OpenACC version of VASP6.3, which runs only on GPU wiki/index.php/OpenACC_GPU_port_of_VASP
Let me know if you need anymore information.

Thank you,
Sophie

[fabien_tran1]

Hi,

Is it only with OpenACC that the problem occurs?

[fabien_tran1]

After a few tries, I could not reproduce this problem. However, the termination occured in your case at the 60th iteration, which may indicate that when the crash occurs may be somehow random and depend sensitively on the setup. If such problems persist, a recommendation would be to use a more recent version of the compiler and/or VASP.