dorsal team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

On 01/09/10 12:52, Anders Logg wrote:
> On Wed, Sep 01, 2010 at 01:42:14PM +0200, Johannes Ring wrote:
>> On Wed, Sep 1, 2010 at 1:30 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> On Wed, Sep 01, 2010 at 12:22:27PM +0100, Garth N. Wells wrote:
>>>>
>>>>
ound. This was changed a few
>>>>>>> days back.
>>>>>>
>>>>>> Did we have CXX=mpic++ or similar in the Dorsal platform file for Snow Leopard?
>>>>>
>>>>> We used to have this in dolfin.package:
>>>>>
>>>>> package_specific_setup () {
>>>>>     export CC=mpicc
>>>>>     export CXX=mpic++
>>>>> }
>>>>>
>>>>> Can we add it back? SLEPc works again for me on OS X when I add it.
>>>>>
>>>>
>>>> Best would be to get the MPI flags right in dolfin/CMakeLists.txt. There
>>>> are a few FIXME related the link flags.
>>>>
>>>> If we use the MPI wrappers, we should somehow test for this and then not
>>>> add the MPI compiler flags in dolfin/CMakeLists.txt.
>>>
>>> I don't understand. That seems like a secondary issue, that we would
>>> add flags we don't really need if we are using the MPI wrappers.
>>>
>>> Does anyone know which extra flag is needed to get things to run?
>>
>> I guess that these should do it: -lmpi -lopen-rte -lopen-pal
> 
> Is there a simple way to find those flags?
> 
>> The problem (in my case with PETSc) is that these are missing from
>> PETSC_LIBRARIES so the test program in FindPETSc.cmake fails.
> 
> Why does PETSc test fail? The PETSc test is run before the SLEPc test
> so any flags added while running the SLEPc test shouldn't affect
> PETSc.
> 
> Or we could add the flags to PETSC_LIBRARIES?
>

We should probably add the MPI flags before testing and configuring the
dependencies.

The Trilinos build system doesn't bother with adding MPI flags - it
relies the on the user to specify the MPI compiler wrappers. Makes the
build system straightforward, but not very user-friendly.

Garth


> --
> Anders