[Pw_forum] DLASCL error on some architectures/QE versions

[Rodrigo Neumann]

Dear QE users and developers,

Not having found any mention to execution-time DLASCL errors in the forum
archives, I though it would be helpful to share my latest experiences with
you. The minimal input file required to run the calculation (and to see the
error) is attached. It's a simple bulk cell calculation with noncollinear
magnetism and spin-orbit coupling using one of the relativistic PPs
provided in the website.

I *sometimes* (the situations in which the error occurs will be detailed
afterwards) get the following error in less than 30 seconds of execution
time:

     Starting wfc are   36 atomic wfcs
 ** On entry to DLASCL parameter number  4 had an illegal value

I'm not at all an expert on BLAS/LAPACK functions but from what I've found
on the online documentation, what DLASCL does is to multiply a M x N real
matrix by a real scalar that is given in the form of a ration CTO/CFROM.
The DLASCL parameter #4 is the double precision variable CFROM which, of
course, should be nonzero. Apparently (both from obvious arithmetic and
the DLASCL manual entry) that is exactly what happens.

What intrigues me is why this error only shows up in specific Architecture
/ QE version pairs...

Situations in which the error ALWAYS shows up:
- IBM/SP6 @ CINECA (AIX 6.1) running QE 4.3.1 with more than 8 processors
- IBM/SP6 @ CINECA (AIX 6.1) running QE 4.2    with more than 8 processors

Situations in which the error NEVER shows up:
- IBM/SP6 @ CINECA (AIX 6.1) running QE 4.1 (tried with up to 32 processors)
- IBM/SP6 @ CINECA (AIX 6.1) running QE 4.3.1 if I set angle1(1)=0 (tried
with up to 32 processors)
- IBM/SP6 @ CINECA (AIX 6.1) running QE 4.2    if I set angle1(1)=0 (tried
with up to 32 processors)
- Argo Linux cluster @ ICTP (CentOS 6) running QE 4.3.2 (tried with up
to 12 processors)
- TMC Linux cluster @ UFRJ (Scientific Linux) running QE 4.1 (tried with up
to 6 processors)

Of course I am already crunching numbers and getting my so-desired results
just by switching to the 4.1 version of pw.x (instead of using the newest
version as I usually do), but then my last 2 days and nights trying to map
the origins of the error would have been useless have I not shared this
with you, with the hope someone might have a clue of what's happening.
Because I don't have the slightest.

Thanks in advance!

With my best regards,
Rodrigo

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

 Rodrigo Neumann Barros Ferreira
Ph.D. Student
Departamento de Física dos Sólidos - Instituto de Física
Universidade Federal do Rio de Janeiro
Cx.Postal 68.528
21945-972, Rio de Janeiro - RJ - BRASIL

Tel: [+55] (21) 2562-7341
http://www.if.ufrj.br/~neumann/
http://tmc.if.ufrj.br/
--------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20120417/aa11f55e/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Co.scf.in
Type: application/octet-stream
Size: 1333 bytes
Desc: not available
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20120417/aa11f55e/attachment.obj>