[Trilinos-Users] trilinos configure failure using pathscale

John R. Cary cary at colorado.edu
Tue Feb 16 18:28:00 MST 2010


FYI to all.  Looks like it built the way I needed.

Thanks!

Another email about BlueGene coming.

John Cary

On 2/16/2010 11:52 AM, Perschbacher, Brent M wrote:
> That is the right variable to be setting, but there is a small caveat 
> to using it in Trilinos. The CMAKE_<compiler>_FLAGS that are set by 
> the user come before our default flags so you won't be able to 
> override anything that we set with that. I don't know that we set 
> anything that conflicts with MPICH_IGNORE_CXX_SEEK, but I can't say 
> for certain until I see a full command line from the build.
>
> Here are the relevant sections of our cmake quickstart file that 
> covers how to add compile flags, or to override all the flags completely.
>
>   b) Adding arbitrary compiler flags but keeping other default flags:
>
>     To append arbitrary compiler flags, configure with:
>
>       -DCMAKE_<COMPILER>_FLAGS:STRING="<EXTRA_COMPILER_OPTIONS>"
>
>     where <COMPILER> = C, CXX, Fortran and <EXTRA_COMPILER_OPTIONS> 
> are your
>     extra compiler options like "-DSOME_MACRO_TO_DEFINE -funroll-loops".
>
>     NOTE: Compiler options set with CMAKE_<COMPILER>_FLAGS will *not* 
> override
>     the set debug or release options which come after on the compile line.
>     Therefore, this approach should only be used for options that will 
> not get
>     overridden by the internally-set debug or release compiler options.
>     However, setting CMAKE_<COMPILER>_FLAGS will work well for adding 
> extra
>     compiler defines for example.
>
>     NOTE: Any options that you set through the cache varible
>     CMAKE_<COMPILER>_FLAGS_<BUILD_TYPE> (where <BUILD_TYPE> = DEBUG or
>     RELEASE) will get overridden in the code so don't try to manually set
>     CMAKE_<COMPILER>_FLAGS_<BUILD_TYPE>.
>
>   c) Overriding debug/release compiler options:
>
>     To pass in compiler options that override the default debug or release
>     options use:
>
>       -D CMAKE_BUILD_TYPE:STRING=NONE \
>       -D CMAKE_C_FLAGS:STRING="-04 -funroll-loops" \
>       -D CMAKE_CXX_FLAGS:STRING="-03 -fexceptions"
>
>     where LANGUAGE = C, CXX, Fortran, etc.
>
>     NOTE: Using CMAKE_BUILD_TYPE:STRING=NONE allows you to take complete
>     control over what debug or release compiler optimizations or related
>     options are passed to the compiler.
>
>
> Brent
>
>
>
>
>
>
> On 2/16/10 10:58 AM, "John R. Cary" <cary at colorado.edu> wrote:
>
>     Thanks.  Got past configure.
>
>     Now trying to get around
>
>     inos-10.0.4p1/par/packages/epetra/src
>     -I/project/projectdirs/vorpal/cary/projects/vpall-franklin/builds-path/trilinos/packages/epetra/src
>
>     -o CMakeFiles/epetra.dir/Epetra_C_wrappers.cpp.o -c
>     /project/projectdirs/vorpal/cary/projects/vpall-franklin/builds-path/trilinos/packages/epetra/src/Epetra_C_wrappers.cpp
>     /opt/cray/xt-asyncpe/3.3/bin/CC: INFO: linux target is being used
>     In file included from
>     /opt/mpt/3.5.0/xt/mpich2-pathscale/include/mpi.h:1110,
>                       from
>     /project/projectdirs/vorpal/cary/projects/vpall-franklin/builds-path/trilinos/packages/epetra/src/Epetra_C_wrappers.cpp:33:
>     /opt/mpt/3.5.0/xt/mpich2-pathscale/include/mpicxx.h:26:2: error:
>     #error
>     "SEEK_SET is #defined but must not be for the C++ binding of MPI"
>     /opt/mpt/3.5.0/xt/mpich2-pathscale/include/mpicxx.h:30:2: error:
>     #error
>     "SEEK_CUR is #defined but must not be for the C++ binding of MPI"
>     /opt/mpt/3.5.0/xt/mpich2-pathscale/include/mpicxx.h:35:2: error:
>     #error
>     "SEEK_END is #defined but must not be for the C++ binding of MPI"
>     make[2]: ***
>     [packages/epetra/src/CMakeFiles/epetra.dir/Epetra_C_wrappers.cpp.o] Error
>     1
>     make[2]: *** Waiting for unfinished jobs....
>
>
>     Do I do
>
>       CMAKE_CXX_FLAGS=-DMPICH_IGNORE_CXX_SEEK
>
>     to have this flag added into any kind of build?  (DEBUG, RELEASE,
>     etc.)...
>
>     Or will this overwrite the other flags?
>
>     Thx....John
>
>
>
>
>
>     On 2/16/2010 8:44 AM, Brad King wrote:
>     > John R Cary wrote:
>     >
>     >> Brad King wrote:
>     >>
>     >>> Please send me (off list) a tarball of the
>     CMakeFiles/FortranCInterface
>     >>> directory from the build tree.
>     >>>
>     >> Here is the full CmakeFiles directory.
>     >>
>     > Try applying the patch below to the CMake Modules directory.
>     >
>     > -Brad
>     >
>     > diff --git a/Modules/FortranCInterface/my_module_.c
>     b/Modules/FortranCInterface/my_module_.c
>     > index 0d7091f..364b214 100644
>     > --- a/Modules/FortranCInterface/my_module_.c
>     > +++ b/Modules/FortranCInterface/my_module_.c
>     > @@ -1,2 +1,8 @@
>     > +#if defined(__PATHSCALE__)
>     > +/* PathScale Fortran wants my_module_ when calling any my_module
>     symbol,
>     > +   but module symbols use '.in.' so we cannot provide them
>     anyway.  */
>     > +void pathscale_my_module_(void) {}
>     > +#else
>     >   /* PGI Fortran wants my_module_ when calling any my_module
>     symbol.  */
>     >   void my_module_(void) {}
>     > +#endif
>     >
>     >
>
>
>     _______________________________________________
>     Trilinos-Users mailing list
>     Trilinos-Users at software.sandia.gov
>     http://software.sandia.gov/mailman/listinfo/trilinos-users
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://software.sandia.gov/pipermail/trilinos-users/attachments/20100216/9f22edd6/attachment.html 


More information about the Trilinos-Users mailing list