Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Brian Barrett (brbarret_at_[hidden])
Date: 2005-09-17 17:46:33


On Sep 17, 2005, at 1:28 PM, Rainer Keller wrote:

> On Friday 16 September 2005 18:41, Brian Barrett wrote:
>
>> A quick question about MPI_Group_range_excl(). I notice that if
>> for a
>> given triplet, first_rank == last_rank, we error if stride is not
>> 1. I
>> notice we don't have that test in MPI_Group_range_incl(). Is there a
>> reason that test should be there - the standard doesn't seem to imply
>> that stride must be 1 if first_rank == last_rank.
>>
> As said in the other mail, the standard claims, that an
> MPI_Group_incl(i) /
> MPI_Group_excl(i) is equivalent to an MPI_Group_range_incl /
> MPI_Group_range_excl with (i,i,1) -- see p. 142 in the MPI-11
> standard.
> So, the test is just for correctness of the supplied data (or
> against lazyness
> of the app.-programmer? ,-])

But it only says that calling MPI_Goup_excl() is equivalent to
calling MPI_Group_range_excl(), with every rank i given to
MPI_Group_excl() becoming a triple (i, i, 1). It says nothing of the
behavior of the function if the triple (i, i, X) if X is not 1...

> But then again, the std. is (as always) not too strict in
> specifying all those
> marginal cases. So, I don't feel too strong about this test.
>
> BTW: the intel-tests failing are in MPI_Group_range_excl3_c.c, right?

correct.

Brian

-- 
   Brian Barrett
   Open MPI developer
   http://www.open-mpi.org/