[Wien] wien2k_04, ifort 8.1, mkl 7.0 on suse 9.1

[Gerhard H Fecher]

Yes I succeded, but not on every computer.

It worked on my laptop (PII 650MHz, 256 MB) running a freshly installed SUSE 
Linux 9.1, I did not succeed on a desktop (P IV 3 GHz, 1.5 MB) running an 
updated Version of SUSE 9.1 (from 8.x to 9.0 and finally to 9.1). In the 
latter case not only Wien2K was crashing with the same symptoms you have, but 
other programs, too.

Actually I do some changes to SUSE 9.2 starting installation from scratch (do 
not use update but a new installation). It seems that some programs are 
running now ok, but I had not yet time to check the latest Version of Wien.

In case you have everything installed natively, did you use dynamic linking ? 
static linking causes troubles with large arrays. 
All arrays have to stay smaller than 2.1GB, otherwise the program crashes. 
Statically linked programs seem also too crash at a certain overall amount of 
large arrays, even if they stay smaller than 2.1GB.
But, don't believe the error messages, they are wrong, even the intel debugger 
fails to find the problem. I have seen that in some cases variables are just 
not passed to subroutines (or with value 0) what often causes an error 
message as the subroutine tries to acess the array element a(0), however, if 
if a(0) is within the allowed dimension, than the program calculates trash 
and you will receive at a complete different place of the program some error 
message, most probably caused by NANs or division by Zeroes. 

It seems that there are incompatibilities with some libraries. First the mkl 
7.0 libraries are not really compatible to the ones coming with ifort 8.1, I 
recommend you may use mkl 7.2. This does not prevent incompatibilities with 
older libraries from glibc that depend on the version of youre Linux 
distribution.

Actually I use:
ifort 8.1.021
mkl 7.2.008
check to have the latest small version, Intel does not provide updates for the 
evaluation versions, so you have to reload the complete packages.

Hope it will help,
Ciao Gerhard




Am Freitag, 25. Februar 2005 11:43 schrieb Lars Lykke:
> Dear WIEN2k users
> 
> Has anyone had a succesfull install of wien2k_04 using the
> - intel ifort 8.1 compiler 
> - intel mkl 7.0 libraries
> - suse 9.1 (or other linux systems?)
> ?
> 
> I can compile easily enough (no obvious compile errors i.e. no errors in
> SRC_lapw2/compile.msg) but if i choose a simple testcase, say Bi, then
> the program will chrash with a segmentation fault when including
> spin-orbit coupling (run_lapw -so).
> init_lapw, lapw0, lapw1 and lapwso seems to run fine.
> 
> If no spin-orbit (i.e. run_lapw) is used the scf-cycle converges ok.
> 
> The last lines in Bi.output2 before the program chrashes with a
> segmentation fault are:
> 
> -------------- Bi.output2 -----------------
> eseper below EF  0.500000000000000
> eseper minimum gap  5.000000000000000E-002
> Energy to separate semicore and valence states: -7.058063866110742E-002
> 
> 
> :NOE  : NUMBER OF ELECTRONS          =  30.000
> 
> :FER  : F E R M I - ENERGY(TETRAH.M.)=   0.33839
> wmist   60.0000000001208
> -------------- Bi.output2 END ----------------
> By comparison with other case.output2, this should be just before the
> force calculation?
> 
> The dayfile reports:
>     start       (Thu Feb 24 18:26:09 CET 2005) with lapw0 (20/20 to go)
> >   lapw0       (18:26:09) 1.517u 0.053s 0:01.57 99.3%  0+0k 0+0io
> 0pf+0w
> >   lapw1       (18:26:11) 10.068u 0.956s 0:11.06 99.5% 0+0k 0+0io
> 0pf+0w
> >   lapwso      (18:26:22) 7.549u 0.419s 0:08.00 99.3%  0+0k 0+0io
> 0pf+0w
> >   lapw2 -c -so        (18:26:30) Segmentation fault
> 0.141u 0.009s 0:00.15 93.3%     0+0k 0+0io 0pf+0w
> 
> >   stop error
> 
> We also experienced compilation problems with the ifort 8.0 compiler
> whereas there have been no problems with the intel ifc 7.1 compiler.
> 
> Any help or pointers are very appreciated
> 
> Best Regards
> Lars Lykke
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>