Re: [Freepooma-devel] Parallel Pooma (Repeated request)

[Richard Guenther]

On Mon, 5 Jun 2006, Svilen Sabchevski wrote:

> Dear Richard,
>
> Thank you for your suggestion to configure with -ex option.
> This indeed solved the current problem and the compilation
> process continued further two more steps. But strangely
> enough after this the problem with SEEK_xxx reappeared
> again compiling BataBrowser.cmpl.cpp. So, I got the following diagnostics:

Ok, for the source level problems with MPI I still think avoiding to
include the mpicxx header at all will cure most of the problems.  F.i.,
in my local MPICH mpi.h header there is the following section:

/* Add the C++ bindings */
/*
   If MPICH_SKIP_MPICXX is defined, the mpi++.h file will *not* be 
included.
   This is necessary, for example, when building the C++ interfaces.  It
   can also be used when you want to use a C++ compiler to compile C code,
   and do not want to load the C++ bindings
 */
#if defined(HAVE_MPI_CPP) && !defined(MPICH_SKIP_MPICXX)
#include "mpi++.h"
#endif

which suggests, that for my particular version/install of MPICH the
problem could be avoided by adding -DMPICH_SKIP_MPICXX (the proper
place to add this is to the config/arch/*.conf file you are using,
just along the optimization flags).

> Probably in the chain of  #includes the order is again wrong and the header 
> mpi.h is
> somehow behind stdio.h. Therefore I decided to experiment with another solution
> of the problems related to SEEK_xxx, namely to add -DMPICH_IGNORE_CXX_SEEK
> in parallel.suite.mk (see next line)
>
> CXX_OPT_LIB_ARGS    = -ftemplate-depth-60    -DNOPAssert -DNOCTAssert -O2 
> -funroll-loops -DMPICH_IGNORE_CXX_SEEK
>
> After this the problem with SEEK_xxx is solved again but  there is still  error 
> message:
>
> Mon Jun 5 16:52:36 EEDT 2006
> Compiler location: /home/mpich2-install/bin/mpiCC
> TMPDIR=/tmp/parallel; \
> /usr/bin/time mpiCC -c src/DataBrowser/DataBrowser.cmpl.cpp \
> -o src/DataBrowser/parallel/DataBrowser.cmpl.o \
>  -ftemplate-depth-60    \
> -DNOPAssert -DNOCTAssert -O2 -funroll-loops -DMPICH_IGNORE_CXX_SEEK \
> -I/home/p_freepooma-2.4.1/src \
> -I/home/p_freepooma-2.4.1/lib/parallel
>
>
>
> In file included from /home/p_freepooma-2.4.1/src/Layout/DynamicEvents.h:55,
>                 from /home/p_freepooma-2.4.1/src/Layout/SparseTileLayout.h:58,
>                 from /home/p_freepooma-2.4.1/src/Engine/IsValidLocation.h:54,
>                 from /home/p_freepooma-2.4.1/src/Array/PrintArray.h:49,
>                 from src/DataBrowser/DataBrowser.cmpl.cpp:23:
> /home/p_freepooma-2.4.1/src/Domain/IteratorPairDomain.h:49: error: redefinition of 
> `struct std::iterator_traits<_Tp*>'
> /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/bits/stl_iterator_base_types.h:138:
>  error: previous definition of `struct std::iterator
> /home/p_freepooma-2.4.1/src/Domain/IteratorPairDomain.h:58: error: redefinition of 
> `struct std::iterator_traits<const _Tp*>'
> /usr/lib/gcc/i686-pc-cygwin/3.4.4/include/c++/bits/stl_iterator_base_types.h:148:
>  error: previous definition of `struct std::iterator

Can you provide the output of the configure command?  Preferably the 
logfile created, too, it's stored in lib/$POOMASUITE/config.log.  Look
for why the test "checking for standard conforming iterators" does fail
and say no.  You can workaround this issue by "fixing" 
lib/$POOMASUITE/PoomaConfiguration.h manually - just fix the define for
POOMA_NONSTANDARD_ITERATOR to POOMA_NO.

> I am sorry for bothering you so frequently. I really appreciate your valuable 
> time.   In an anticipation of
> your next hint I would like to acknowledge that each your response helps me to 
> make several more
> steps further.

I know getting it to work in parallel is not easy, but we may be able
to improve the autodetection to avoid some of the common pitfalls.

Richard.

--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/