The ABINIT Users Mailing List ( CLOSED )

["Razvan Caracas" <r.caracas@gl.ciw.edu>]

The problem seems to be quite serios and I wonder if we should not move
the discussion on the developer's list.

It seems to me that it comes from the fact that there are zero-sized
arrays in the code. For example, in case of a non-alchemical calculation,

ntypalch=0
npspalch=0 (<= =npsp - ntyppure, which is ntypat, so npsp)

in */driver.F90,
  allocate(dtset%mixalch(npspalch,ntypalch))
will generate a mixalch array with dimension 0x0.

fortran 90 allows zero-sized arrays, and even operations, like addition,
loop into, etc. with these arrays (even if some f90 compilers still have
bugs), the c and c++ compilers do not accept zero-sized arrays.

The code crashes only when dtset%mixalch is mailed to */getdim_nloc.F90
within */driver.F90.

So if the above problem can be solved relatively easy with an if statement
avoiding this zero-sized allocation, further away in the code in
*/scfcv.F90   the psps%ncspl is mailed to */stvtr.F90 (and others) and the
code crashes again during execution, because during a non-PAW calculation
this matrix is again zero-sized. It's just that now the problem is more
complex in terms of side-effects and routines to be modified.

So briefly, are there any c or c++ compilers that support zero-sized
arrays? Or some flags that would make them to support these arrays? Or the
only way to avoid this is just to go and add if statements? How comes then
that the same code works on other machines or with other compilers?

Many thanks,
Razvan

PS The problem is relatively old, as versions 4.5 and 4.4 behave
similarly.

> Le jeudi 19 janvier 2006 à 16:52 +0100, r.caracas@gl.ciw.edu a écrit :
>
>> Dear all,
>> I try to compile abinit-4.6.5 on an opteron cluster with pgf90 compiler
v-6.0 for linux-64. I tried different flags but every time the code
compiles but crashes at the same point EVERY time.
>> [...]
>> This happens with both sequential and parallel version of the code.
Obviously, none of the tests works.
>> Anybody has any suggestions of what might be wrong?
>
> We get the same problem with ABINIT 5.0, and are currently
> investigating. Though the glibc message is already a hint, it is sure
that any help will be greatly appreciated.
>
> Best regards,
>
> Yann.
>
> --
> Yann Pouillon                      UCL/SE/FSA/MAPR/PCPM
> Tel: +32 (0)10 47 81 91            Bâtiment Boltzmann, a+1
> Fax: +32 (0)10 47 34 52            1 place Croix du Sud
> Mel: pouillon@pcpm.ucl.ac.be       1348 Louvain-la-Neuve (Belgique)
>


-- 
===============================================
Dr. Razvan Caracas
PostDoctoral Fellow
Carnegie Institution of Washington,
Geophysical Laboratory,
5251 Broad Branch Rd. N.W.,
Washington DC 20015, USA
tel: 1-202-478-8945; fax: 1-202-478-8901
http://www.gl.ciw.edu/~r.caracas