[Mpi-22] [MPI Forum] #109: Nonblocking Collective Operations

[Bronis R. de Supinski]

Rolf and Torsten:

It really seems to me that these changes exceed
the guidelines that Rich and Bill have given
for the release draft. While I understand the
reasoon behind them, it is unclear that we will
have reviewed them properly. For example, I think
the wording below could use work.

I am in transit to ANL right now so I cannot go
over it properly until at least tonight...

Even then, this feels rushed.

Bronis

On Tue, 2 Nov 2010, MPI Forum wrote:

> #109: Nonblocking Collective Operations
> --------------------------------------------------------+-------------------
>                    Reporter:  htor                     |                       Owner:  htor
>                        Type:  New routine(s)           |                      Status:  new
>                    Priority:  Waiting for PDF reviews  |                   Milestone:  2009/02/09 San Jose
>                     Version:  MPI 3.0                  |                    Keywords:
>              Implementation:  Completed                |           Author_bill_gropp:  1
>          Author_rich_graham:  0                        |           Author_adam_moody:  0
>      Author_torsten_hoefler:  1                        |        Author_dick_treumann:  0
> Author_jesper_larsson_traeff:  0                        |       Author_george_bosilca:  0
>           Author_david_solt:  0                        |   Author_bronis_de_supinski:  0
>        Author_rajeev_thakur:  0                        |         Author_jeff_squyres:  0
>    Author_alexander_supalov:  0                        |    Author_rolf_rabenseifner:  1
> --------------------------------------------------------+-------------------
>
> Comment(by htor):
>
> Replying to [comment:44 RolfRabenseifner]:
> > Whether we use now the text in my last comment or, a more complete text,
> > we can decide now or later, but I would prefer a more complete text,
> > because then it is clear why this exception is only for 4 completion
> routines:
> >
> >   The fields in a status object returned by a call to MPI_WAIT,
> MPI_TEST, or
> >   MPI_{TEST|WAIT}{ALL|SOME|ANY}, where the request corresponds to a
> nonblocking
> >   collective call, are undefined, with one exceptions: The error status
> field
> >   will contain valid information if the wait or test call returned with
> MPI_ERR_IN_STATUS.
> >   MPI_ERR_IN_STATUS can be returned only by MPI_{TEST|WAIT}{ALL|SOME},
> because
> >   only these MPI completion functions take arrays of MPI_STATUS.
>
> Rolf and I decided to adopt this minor change (with slight modifications)
> and clear fix after a phone call. It now reads:
>
> "The fields in a status object returned by a call to MPI_WAIT, MPI_TEST,
> or any of the other derived functions (MPI_{TEST|WAIT}{ALL|SOME|ANY}),
> where the request corresponds to a send call, are undefined, with one
> exception: The error status field will contain valid information if the
> wait or test call returned with MPI_ERR_IN_STATUS. MPI_ERR_IN_STATUS
> should be returned only by MPI_{TEST|WAIT}{ALL|SOME}, because only these
> MPI completion functions take arrays of MPI_STATUS."
>
> See revision 679.
>
> Thanks,
>    Torsten
>
> -- 
> Ticket URL: <https://BLOCKEDsvn.mpi-forum.org/trac/mpi-forum-web/ticket/109#comment:47>
> MPI Forum <https://BLOCKEDsvn.mpi-forum.org/>
> MPI Forum