[MPIWG Fortran] ticket 351
jeff.science at gmail.com
Tue Oct 22 10:44:16 CDT 2013
I am hesitant to assume this problem is resolved just because all
Linux-based clusters are fine with this*. I don't think that this
will be compiler sensitive either, so testing 5 compilers doesn't
prove much. Testing 5 process launchers might.
* This is not meant to disparage Cray by suggesting their machines
behave like Linux clusters in terms of the process environment;
rather, this is a good thing.
On Tue, Oct 22, 2013 at 10:30 AM, Bill Long <longb at cray.com> wrote:
> Hi Jeff,
> I admit that the Fortran standard is (intentionally) vague about what
> constitutes a "command line". However, I tried the 5 compilers I have
> available (Cray, Intel, PGI, gfortran, Pathscale) with a simple program that
> prints argument 0 == the "command name by which the program was invoked",
> and in all cases the launcher text (aprun -n1 ) was ignored and the program
> printed "./a.out". Not conclusive, but suggests that the vendors have
> figured out what the right "quality of implementation" action is here. If
> there are examples to the contrary, I'd be interested in knowing about them.
> On 10/22/13 10:12 AM, N.M. Maclaren wrote:
>> On Oct 22 2013, Bill Long wrote:
>>> 2) I don't really see the point of this at all. The user can just
>>> call the native Fortran intrinsics directly. Why should there be a
>>> duplicate version that starts with MPI_ ? I would suggest, instead,
>>> that you just add a note in the spec (advice to users) point out that
>>> Fortran programmers can access the command line arguments using the
>>> language intrinsics. Maybe someone can convince me this is not
>>> adequate, but I've been using those intrinsics with parallel programs
>>> for some time and they work just fine.
>> I assume that the desire for this is to support the systems which do not
>> use a mpiexec/mpirun command, but fire up MPI directly, as IBM poe does.
>> However, I have always felt that the original C approach was a horrible
>> hack, just as having C's argument list (and arguments) writable is.
>> There were good reasons at the time but, given that Fortran has done
>> without it with little to no trouble for so long, I agree with Bill.
>> Nick Maclaren.
>> mpiwg-fortran mailing list
>> mpiwg-fortran at lists.mpi-forum.org
> Bill Long longb at cray.com
> Fortran Technical Support & voice: 651-605-9024
> Bioinformatics Software Development fax: 651-605-9142
> Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101
> mpiwg-fortran mailing list
> mpiwg-fortran at lists.mpi-forum.org
jeff.science at gmail.com
More information about the mpiwg-fortran