dorsal team mailing list archive

[Harish Narayanan <harish.mlists@xxxxxxxxx>]

On 8/26/10 2:11 PM, Harish Narayanan wrote:
> On 8/26/10 2:01 PM, Garth N. Wells wrote:
>> We should think again about using MacPorts MPI on OSX. If we use the
>> MacPorts MPI, I expect that
>>
>>  - We can build PETSc with MUMPS (and any other Fortran packages)
>>  - We can use CMake FindFoo.cmake for BLAS and LAPACK
>>  - We won't have to worry about defects of older compilers
> 
> I thought about this after it came up the last time, then realised I am
> not sure how to do this. MacPorts creates gfoo-mp-4.4 compilers for
> various languages, but its OpenMPI points to the system compilers
> (gfoo-4.2).
> 
> MacPorts has a 'variants' option while installing packages to fine tune
> things,
> 
> openmpi has the variants:
>    g95: build mpif77 and mpif90 using g95
>      * conflicts with gcc42 gcc43 gcc44 gcc45
>    gcc42: build mpif77 and mpif90 using gcc42
>      * conflicts with g95 gcc43 gcc44 gcc45
>    gcc43: build mpif77 and mpif90 using gcc43
>      * conflicts with g95 gcc42 gcc44 gcc45
> [+]gcc44: build mpif77 and mpif90 using gcc44
>      * conflicts with g95 gcc42 gcc43 gcc45
>    gcc45: build mpif77 and mpif90 using gcc45
>      * conflicts with g95 gcc42 gcc43 gcc44
>    progressthreads: enable threads asynchronous communication progress
>    threads: enable threads for MPI applications
> 
> though I am not sure how to read its output for OpenMPI to make it do
> what you're asking for (i.e. point to a more recent MacPorts gfoo).

Forget what I said, I think I was doing something stupid and confusing
MacPorts' mpifoo with the system mpifoo. I will try to make this change
and report how it goes.

Harish