[mpich-discuss] MPI comm attributes not working?
Jeff Hammond
jeff.science at gmail.com
Tue Apr 5 14:38:40 CDT 2016
Thanks, Ken. When I set *flag=1 in the copyfn, the output becomes:
$ ./a.out
In ibcast_copy: keyval = -1539309568
In ibcast_delete: keyval = -1539309568
In ibcast_delete: keyval = -1539309568
Jeff
On Tue, Apr 5, 2016 at 11:08 AM, Kenneth Raffenetti <raffenet at mcs.anl.gov>
wrote:
>
> I think in the example, it's undefined if the attribute will be set on
the duped communicator. See pg. 269 in MPI-3.1. If the callback does not
set the output flag to true/false, how will the implementation know what to
do?
>
> Ken
>
> On 04/05/2016 11:05 AM, Jeff Hammond wrote:
>>
>> I guess I don't know what the right answer is for MPI_COMM_WORLD. I'll
>> try that test again with attributes set only on subcomms to see how it
>> behaves.
>>
>> Jeff
>>
>> On Tue, Apr 5, 2016 at 10:24 AM, Balaji, Pavan <balaji at anl.gov
>> <mailto:balaji at anl.gov>> wrote:
>>
>>
>> Hmm. I would think the mpich output is correct: the attributed
>> should only be deleted once. But I could be wrong.
>>
>> -- Pavan
>>
>> > On Apr 5, 2016, at 8:10 AM, Jeff Hammond <jeff.science at gmail.com
>> <mailto:jeff.science at gmail.com>> wrote:
>> >
>> > My concern is the number of times the string "ibcast_delete"
>> appears in the output (2 vs 1)...
>> >
>> > Jeff
>> >
>> > On Mon, Apr 4, 2016 at 10:02 PM, Balaji, Pavan <balaji at anl.gov
>> <mailto:balaji at anl.gov>> wrote:
>> >
>> > You are printing a number that's generated by the MPI
>> implementation. It's value will be different for different MPI
>> implementations.
>> >
>> > -- Pavan
>> >
>> >> On Apr 4, 2016, at 11:56 PM, Jeff Hammond
>> <jeff.science at gmail.com <mailto:jeff.science at gmail.com>> wrote:
>> >>
>> >> I am trying to understand how communicator attributes work, so I
>> found
>>
http://stackoverflow.com/questions/25198353/callbacks-provided-in-mpi-comm-create-keyval-are-not-called
,
>> which contains a nice example (the code in
>> http://stackoverflow.com/a/25205611/2189128).
>> >>
>> >> However, the posted output of this code is:
>> >>
>> >> $ mpirun -np 1 ./comm-attr
>> >> In ibcast_copy: keyval = 10
>> >> In ibcast_delete: keyval = 10
>> >> In ibcast_delete: keyval = 10
>> >>
>> >> Whereas I observe:
>> >>
>> >> $ mpirun -np 1 ./a.out
>> >> In ibcast_copy: keyval = -1539309568
>> >> In ibcast_delete: keyval = -1539309568
>> >>
>> >> Can anyone help me understand what the correct output of this
>> program is? If it is not what I am seeing, is this a bug in MPICH
>> or am I doing something wrong?
>> >>
>> >> I am using MPICH git from November 2015, so I will git pull &&
>> ./autogen.sh && $CONFIGUREMAKEINSTALL and try again just to be sure.
>> >>
>> >> Thanks,
>> >>
>> >> Jeff
>> >>
>> >> $ mpirun --version
>> >> HYDRA build details:
>> >> Version: 3.2
>> >> Release Date: unreleased
>> development copy
>> >> CC: icc
>> >> CXX: icpc
>> >> F77: ifort
>> >> F90: ifort
>> >> Configure options:
>> '--disable-option-checking'
>> '--prefix=/opt/mpich/dev/intel/default' 'CC=icc' 'CXX=icpc'
>> 'FC=ifort' 'F77=ifort' '--enable-cxx' '--enable-fortran'
>> '--enable-threads=runtime' '--enable-g=dbg' '--with-pm=hydra'
>> '--enable-wrapper-rpath' '--enable-static' '--enable-shared'
>> '--cache-file=/dev/null' '--srcdir=../../../../src/pm/hydra'
>> 'CFLAGS= -g -O2' 'LDFLAGS=' 'LIBS=-lpthread ' 'CPPFLAGS=
>> -I/Users/jrhammon/Work/MPI/MPICH/git/build/src/mpl/include
>> -I/Users/jrhammon/Work/MPI/MPICH/git/src/mpl/include
>> -I/Users/jrhammon/Work/MPI/MPICH/git/src/openpa/src
>> -I/Users/jrhammon/Work/MPI/MPICH/git/build/src/openpa/src
>> -D_REENTRANT
>> -I/Users/jrhammon/Work/MPI/MPICH/git/build/src/mpi/romio/include'
>> >> Process Manager: pmi
>> >> Launchers available: ssh rsh fork slurm
>> ll lsf sge manual persist
>> >> Topology libraries available: hwloc
>> >> Resource management kernels available: user slurm ll lsf
>> sge pbs cobalt
>> >> Checkpointing libraries available:
>> >> Demux engines available: poll select
>> >>
>> >>
>> >> --
>> >> Jeff Hammond
>> >> jeff.science at gmail.com <mailto:jeff.science at gmail.com>
>> >> http://jeffhammond.github.io/
>> >> _______________________________________________
>> >> discuss mailing list discuss at mpich.org <mailto:discuss at mpich.org>
>> >> To manage subscription options or unsubscribe:
>> >> https://lists.mpich.org/mailman/listinfo/discuss
>> >
>> > _______________________________________________
>> > discuss mailing list discuss at mpich.org <mailto:discuss at mpich.org>
>> > To manage subscription options or unsubscribe:
>> > https://lists.mpich.org/mailman/listinfo/discuss
>> >
>> >
>> >
>> > --
>> > Jeff Hammond
>> > jeff.science at gmail.com <mailto:jeff.science at gmail.com>
>> > http://jeffhammond.github.io/
>> > _______________________________________________
>> > discuss mailing list discuss at mpich.org <mailto:discuss at mpich.org>
>> > To manage subscription options or unsubscribe:
>> > https://lists.mpich.org/mailman/listinfo/discuss
>>
>> _______________________________________________
>> discuss mailing list discuss at mpich.org <mailto:discuss at mpich.org>
>> To manage subscription options or unsubscribe:
>> https://lists.mpich.org/mailman/listinfo/discuss
>>
>>
>>
>>
>> --
>> Jeff Hammond
>> jeff.science at gmail.com <mailto:jeff.science at gmail.com>
>> http://jeffhammond.github.io/
>>
>>
>> _______________________________________________
>> discuss mailing list discuss at mpich.org
>> To manage subscription options or unsubscribe:
>> https://lists.mpich.org/mailman/listinfo/discuss
>>
> _______________________________________________
> discuss mailing list discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
--
Jeff Hammond
jeff.science at gmail.com
http://jeffhammond.github.io/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpich.org/pipermail/discuss/attachments/20160405/651bd18d/attachment.html>
-------------- next part --------------
_______________________________________________
discuss mailing list discuss at mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
More information about the discuss
mailing list