The ABINIT Users Mailing List ( CLOSED )

[James Johns <jej7x@xenon.cchem.berkeley.edu>]

Thanks again Alain.  Ok, so I copmiled gcc, g++ gfortran all version 4.3, 
and then rebuilt openmpi.  The ./configure script worked fine, but then 
make failed again and again, regardless of the compiler flags I passed it. 
I did succesfully compile the 5.6.5 parallel version, but to do that I had 
to compile openmpi using mixed compiler suites:  gcc, g++,  and ifort. 
Obviously, this is not the most desireable situation and to be honest i'm 
shocked that openmpi even compiled.  This time while running ./configure 
with the usual flags and FC=ifort, it compiled and I didn't see any bugs 
in make, but it did fail during "make install" while making the test 
directories.  I'm currently running a test calculation for a 7 atom thick 
Ag slab optimizing the unit cell size, kpt grid, tsmear etc, and comparing 
it previously run serial calculations from version 5.4.  Is there any 
thing in particular I should be on the look out for, and do you have any 
final ideas why I wouldn't have been able to build abinit using gfortran 
4.3.3 and openmpi?
James

On Tue, 17 Feb 2009, Alain Jacques wrote:

> Hello James,
>
> Glad to know that you're going in the right direction.
>
> Excerpt from the
> http://www.abinit.org/Infos_v5.6/known_problems/KNOWN_PROBLEMS file ...
> a recommended reading :-)
>
> "Versions of compilers known to be too buggy for ABINIT
> gfortran 4.1.2    (=> use gfortran 4.2 or 4.3 instead)
> intel10.0         (=> use intel9.1 or 10.1 instead)"
>
> So I am afraid your gcc/gfortran is outdated.
> You have two solutions:
>
> * you earlier mentioned that you have an Intel ifort installed ... I use
> recent subversions of 10.1 (supported) and 11.0 (no apparent problem).
> OpenMPI compiles nicely with Intel ifort/icc - setup your environment
> and configure openMPI with --enable-shared --enable-static CC=icc
> F77=ifort FC=ifort CXX=icpc options (use --prefix=... to install it in a
> nice place)
>
> * or compile a new gcc/gfortran toolchain with your built in gcc-4.1.2
> - easy as pie. Download gcc-4.3.2 or gcc-4.3.3 from http://gcc.gnu.org/,
> - read the installation doc! -, configure with  --enable-bootstrap
> --enable-shared --enable-threads=posix --enable-checking=release
> --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions
> --enable-languages=c,c++,fortran --with-cpu=generic
> --build=x86_64-suse-linux options (and use --prefix=)
> recompile openmpi for this new gcc/gfortran toolchain ... similar recipe.
>
> Don't forget to export the right PATH and LD_LIBRARY_PATH to the right
> compilers execs and libraries (check 32/64bit paths) in order to have
> the new tools invoked before the default gcc/gfortran/mpi stuff.
> Reconfigure abinit to reflect the new toolchain - examples available in
> http://www.etsf.be/downloads/ETSF_softlib/page2/page2.html. Build abinit
> with a plain "make" - "make multi" only tries a parallel build for
> speed, doesn't work any better.
>
> The "conftest.c:16:28: error: ac_nonexistent.h: No such file or
> directory" are benign. Your module problem is probably a compiler bug -
> you may try to workaround it by specifying an include path (beurk!).
>
>
> Kind regards,
>
> Alain
>
>
> James Johns wrote:
> > Thanks for the help Alain, i took your advice and installed openmpi
> > into /opt/openmpi and exported the paths as you said. I then compiled
> > the simple hello world program below using mpifortran and ran the
> > a.out using "mpirun a.out" and it worked just fine.  I then switched
> > to my abinit directory  to try configure --enable-mpi
> > --with-mpi-prefix=/opt/openmpi.  I thought the configuration script
> > ran fine, and tried make multi and make multi_alt, but I kept getting
> > this error.
> ...