Linux - IFC 10.1 Abinit automatic test fail

sestak · Post by **sestak** » Fri May 14, 2010 2:27 pm

Dear Abinit users,

I successfully compiled the Abinit serial version (6.0.3) using IFORT, ICC, ICPC Version 10.1 compiler with optimalization level 2 on linux machine. After that, I ran the test in "test" directory using the "make tests_min4" command. However, when I looked to the report files in the test subdirectories I found some errors in dir v4-6. Most of these errors look similar so I describe only the errors in the v4 dir here.
The report files contain these warning:

v4:
======================================================================
Case_84
The diff analysis cannot be done : the number of lines to be analysed
differ.
File 1: 1505 lines, 55 ignored
File 2: 1506 lines, 55 ignored

Summary Case_84 : fatal error, see above message

Case_85
The diff analysis cannot be pursued : the command sign differ.
File 1, line 1636, 55 ignored, character:
File 2, line 1636, 55 ignored, character:=

Summary Case_85 : fatal error, see above message

Case_86
The diff analysis cannot be pursued : the command sign differ.
File 1, line 924, 57 ignored, character:=
File 2, line 924, 57 ignored, character:

Summary Case_86 : fatal error, see above message

Case_87
The diff analysis cannot be pursued : the command sign differ.
File 1, line 1121, 59 ignored, character:=
File 2, line 1121, 59 ignored, character:

Summary Case_87 : fatal error, see above message

Case_88
The diff analysis cannot be done : the number of lines to be analysed
differ.
File 1: 1005 lines, 56 ignored
File 2: 1006 lines, 56 ignored

Summary Case_88 : fatal error, see above message
======================================================================

Detailed examination of diff files (diff.t8x) shows, that significant differences between my out files and ref. out files can be found. The diff files related to these error are included. (I cant upload the files. So, the output can be found: http://jaja.kn.vutbr.cz/~sestak/)

I would like to ask if there is a way how to solve this problem or I can ignore it. I was also trying to find some answers in this forum and I have found the thread of Prof. Zwanziger at link:
http://forum.abinit.org/viewtopic.php?f=18&t=55&p=160&hilit=the+command+sign+differ&sid=6b9e57daad09a60643c8d3dc782c9456#p160.
but I don't know if this solution will help me. If some other files will be needed I will send them.

Thanks for suggestions

Best Regards

Petr Sestak

--
Dr. Petr Sestak
Institute of Engineering Physics
Faculty of Mechanical Engineering
Brno University of Technology
Technicka 2896/2, Brno 616 69, Czech Republic

Tel.: +420 541142812
Mail: sestakATfmeDOTvutbrDOTcz

gmatteo · Post by **gmatteo** » Fri May 14, 2010 5:59 pm

Dear Sestak,

I had a look at the output files of the failing tests. All of them concern GW calculations (the features added in v4).
It seems that the direct diagonalization of the KS Hamiltonian produces completely wrong results leading to Nans
in the GW part.
At this point I think there's some problem in the blas or lapack installation but we need more information
on the compilation options. Are you using the internal linalg library or an external blas-lapack library.
Could you upload your config.log file as well as the the log files of the failing tests so that we can have a look at the problem in more detail.

Thanks for reporting the problem,
Matteo

sestak · Post by **sestak** » Sat May 15, 2010 12:29 pm

Hi Matteo,
Thanks for your prompt reaction. I added all requested files into link http://www.kn.vutbr.cz/~sestak. There you can find my makefile "localhost.ac" as well as configure.log with all logs files.

Regards
Peter

sestak · Post by **sestak** » Mon May 17, 2010 10:16 pm

Hi Matheo,

thanks again. I compiled the Abinit with an external blas and lapack libraries. I used standard blas and lapack in our linux distribution. The makefile was completely same as I used before. After that, I ran the test again and the most of the mentioned errors disappeared. However, four errors still remain. Two of them are in dir v5 a two in dir v6.

v5 t77:

Code: Select all

 Case_77
2
< .Version 6.0.3  of ABINIT 
> .Version 6.0.0  of ABINIT 
3
< .(sequential version, prepared for a i686_linux_intel10.1 computer) 
> .(MPI version, prepared for a x86_64_linux_gnu4.4 computer) 
17
< .Starting date : Mon 17 May 2010.
> .Starting date : Mon 18 Jan 2010.
1113
<  Fermi (or HOMO) energy (hartree) =   0.22573   Average Vxc (hartree)=  -0.37996
>  Fermi (or HOMO) energy (hartree) =   0.23035   Average Vxc (hartree)=  -0.37996
2457
<  Fermi (or HOMO) energy (hartree) =   0.22575   Average Vxc (hartree)=  -0.37997
>  Fermi (or HOMO) energy (hartree) =   0.23036   Average Vxc (hartree)=  -0.37997
2759
< +Overall time at end (sec) : cpu=        182.3  wall=        193.3
> +Overall time at end (sec) : cpu=         68.9  wall=         68.9
Summary Case_77 : different lines= 2 , max discrepancies absolute= 0.00462000000000001 , relative= 0.0101298017891598

Code: Select all

v5 t80:
 Case_80
The diff analysis cannot be pursued : the command sign differ.
File 1, line 897, 103 ignored, character:=
File 2, line 897, 103 ignored, character: 

Summary Case_80 : fatal error, see above message

The last two errors (t21 and t91) are similar to the others. So, I did not describe them in detail here. All mentioned failed tests can be found in the out files at the link http://www.kn.vutbr.cz/~sestak/compilation_with_external_lapack_and_blas. I dont know what to do with these errors. The two of them are due to some differences between ref and my values and I am not sure if these differences are important or not. I will appreciate any suggestion how to solve this.

Best Regards

Peter

ABINIT Discussion Forums

Linux - IFC 10.1 Abinit automatic test fail

Linux - IFC 10.1 Abinit automatic test fail

Re: Linux - IFC 10.1 Abinit automatic test fail

Re: Linux - IFC 10.1 Abinit automatic test fail

Re: Linux - IFC 10.1 Abinit automatic test fail