The ABINIT Users Mailing List ( CLOSED )

[Mustafa Uludogan <uludoganmustafa@yahoo.com>]

Hi,
Using -qextname in xlf causes problems especially with
the parallel version of abinit. Abinit group also
advises not to use -qextname with the parallel
installation. On MAC systems, if you want to get full
advantage of VecLib , you have to use -qextname, but
from my experience using   other codes on MAC , I do
not prefer to use -qextname flag. I always have two
versions of lam or mpich installed with -qextname and
without it.
Mustafa
--- Fabrizio Cleri <fabrizio.cleri@casaccia.enea.it>
wrote:

> Hi,
> 
> there is nothing wrong in the makefiles, in fact the
> ones 
> Mustafa sent me were pretty close to the ones I had.
> 
> The problem is in the "underscore" character _ which
> is to 
> be appended, or not, to some fortran libraries when
> mixing 
> fortran and c calls. The fact is that all the mpi 
> libraries do not have the underscore, by default,
> while 
> some libraries necessary to Abinit (usually the BLAS
> and 
> some other stuff) do need the _ character. 
> 
> The only way around I could find is to compile also
> the 
> mpi libraries by adding the _ suffix. This can be
> done by 
> setting FFLAGS and F90FLAGS to include the -qextname
> 
> option.
> 
> When such libraries are produced, the link step is
> done 
> correctly. However, there are additional problems
> which I 
> have found. 
> 
> For example, if I simply compile the mpich-1 library
> from 
> Dauger Research (the one compatible with Pooch)
> adding the 
> -qextname, the libraries are produced, but some
> commands, 
> including mpif90 which is necessary to make Abinit,
> come 
> out wrong. So I had to do TWO compilations, one with
> 
> -qextname to get the libraries right, and another
> without 
> -qextname to get all the commands right. Now I mix
> the two 
> and Abinit compiles well.
> 
> Then, I have some problems to start the parallel 
> compilation. Apparently, abinip starts and works
> with 1 
> process. But if I start more than 1 process the
> whole 
> thing crashes. However, the mpi tests run well and
> also 
> the Pooch sample programs work well... 
> 
> still to clarify...
> 
> Keep me posted about your progress,
> fc
> 
> 


Mustafa Uludogan
Chemical Engineering Department
Texas A&M University
College Station, TX 77843


                
____________________________________________________
Start your day with Yahoo! - make it your home page 
http://www.yahoo.com/r/hs