SIGSEGV, segmentation fault

Message

rudy.coquet · #1 Post by **rudy.coquet** » Tue Feb 14, 2006 4:19 am

Dear VASP users,

I have compiled VASP 4.6.26 on a P4 Linux cluster (8 cpus) using ifc 9.0 and mkl 8.0. This is the second 8 cpu cluster onto which I compile VASP and I used exactly the same options, compilers and libraries for both clusters. However, the new cluster is installed with RHEL 4 (corresponding to Fedora Core 3) while the "old" one is a RHL 7.2 (corresponding to RHEL 2.1). I am also using LAM-MPI.
On the recently installed cluster, when running VASP (serial but parallel gives the same kind of error), everything starts normally but finally VASP crashes with:

"forrtl: severe (174): SIGSEGV, segmentation fault occurred stack trace terminated abnormally."

The only thing written to the stdout is:

"vasp.4.6.26 15Dec04 complex
POSCAR found : 2 types and 6 ions
LDA part: xc-table for Pade appr. of Perdew
POSCAR, INCAR and KPOINTS ok, starting setup
FFT: planning ... 1
reading WAVECAR"

and the tail of the OUTCAR is:

"initial charge density was supplied:
charge density of overlapping atoms calculated
number of electron 44.0000000 magnetization
keeping initial charge density in first step

----------------------------------------------------------------------------------------------------
----

Maximum index for augmentation-charges 10845 (set IRDMAX)"

There is no other information...
I installed ifc and mkl as non-root AND as root: same result.
I also tried to recompile VASP with less agressive optimisation flags: again, same thing.
However, if I simply copy my 'vasp' executable file from my first cluster to the new one, everything works fine!

I don't understand at this point what is going on and would really appreciate your help with this.

With best regards,
Rudy C.

avan · #2 Post by **avan** » Thu Jan 25, 2007 8:42 am

This problem appears over and over again:(
It seems to appear not in all cases, but I haven't caught any orderliness.
There are several presumption but I am not sure:
1.x86_64 (I encountered this kind of problem only on the x86_64 linux);
2.lib (libgoto in my case);

Can anybody give me some advice?
Thank afore.

Avan.

#3 Post by **admin** » Mon Jan 29, 2007 2:11 pm

please first of all upgrade vasp to the latest release (4.6.28) and check whether the error persists.

avan · #4 Post by **avan** » Tue Jan 30, 2007 5:16 am

[quote="admin"]please first of all upgrade vasp to the latest release (4.6.28) and check whether the error persists.[/quote]

yes. the one I used is the latest release (4.6.28). And the error does exists when the libgoto is used in my cases.

When I changed libgoto to Intel mkl 9.0, some cases which are failed when libgoto is used are performed normally.

So, maybe it is due to the imperfectness of libgoto on x86_64 OS. I'm not sure. It is just a guess based on the tests I have done. It should be noticed that I have only tested two version (r1.02 and r1.09) of libgoto. They seems not perfect for the VASP calculation on x86_64 OS. Certainly, there is another possibility that I have made improper compile settings on libgoto. (I just used the quickbuild.64bit script to build the lib in the case of r1.09 version.)

#5 Post by **admin** » Tue Jan 30, 2007 4:29 pm

probably your lobgoto is not built correctly, you are right. We have never had similar troubles with this library