[Mpi3-rma] Fwd: [Mpi-comments] Freeing memory allocated in MPI_WIN_ALLOCATE_SHARED

[Jeff Hammond]

This must be fixed because it is impossible for the user to free
shared memory in a portable way.  To not have this text is to create
memory leaks in every code that calls MPI_WIN_ALLOCATE_SHARED.

I find that POSIX implementations are far from standard in terms of
(de)registration/(de)allocation/truncation.  Linux, Mac and Blue Gene
do not behave the same way, which is quite obnoxious.

Jeff

On Wed, Sep 5, 2012 at 9:50 AM, William Gropp <wgropp at illinois.edu> wrote:
> Here's the current text (p 415 line 6):
>
> If the
> window was created with \mpifunc{MPI\_WIN\_ALLOCATE},
> \mpifunc{MPI\_WIN\_FREE} will free the window memory that was allocated
> in \mpifunc{MPI\_WIN\_ALLOCATE}.}
>
> Here is my proposed correction
>
> (above text, followed by)
>
> If the window was created with \mpifunc{MPI\_WIN\_ALLOCATE\_SHARED},
> \mpifunc{MPI\_WIN\_FREE} will free the window memory that was allocated
> in \mpifunc{MPI\_WIN\_ALLOCATE\_SHARED}.
>
> This is a sufficient change, since the document is a standard and we don't need to state the same behavior everywhere (note that MPI_WIN_ALLOCATE doesn't mention MPI_WIN_FREE in the description of MPI_WIN_ALLOCATE).
>
> Thoughts?  Is this a "simple oversight" error that we can add?
>
> Bill
>
>
> William Gropp
> Director, Parallel Computing Institute
> Deputy Director for Research
> Institute for Advanced Computing Applications and Technologies
> Paul and Cynthia Saylor Professor of Computer Science
> University of Illinois Urbana-Champaign
>
>
>
> On Sep 5, 2012, at 12:39 AM, Jeff Hammond wrote:
>
>> Okay, I thought it might do that but despite having a paper copy of
>> the MPI-3 RMA chapter three feet from me at the time, I was too lazy
>> to look it up.
>>
>> Thanks for the clarification.
>>
>> Jeff
>>
>> On Tue, Sep 4, 2012 at 11:47 PM, Jim Dinan <dinan at mcs.anl.gov> wrote:
>>> MPI_Win_free also frees memory allocated by MPI_Win_allocate.
>>>
>>> ~Jim.
>>>
>>>
>>> On 9/4/12 4:22 PM, Jeff Hammond wrote:
>>>>
>>>> Shouldn't that be a new call (MPI_Win_free_shared) to keep the usage
>>>> of MPI_Win_free uniform?  MPI_Win_free doesn't deallocate the buffers
>>>> in any other situation, does it?
>>>>
>>>> Jeff
>>>>
>>>> On Tue, Sep 4, 2012 at 4:07 PM, Jim Dinan <dinan at mcs.anl.gov> wrote:
>>>>>
>>>>> Hi All,
>>>>>
>>>>> It looks like we forgot to include the text specifying that
>>>>> MPI_Win_allocate_shared memory is freed along with the window.
>>>>>
>>>>>  ~Jim.
>>>>>
>>>>>
>>>>> -------- Original Message --------
>>>>> Subject:        [Mpi-comments] Freeing memory allocated in
>>>>> MPI_WIN_ALLOCATE_SHARED
>>>>> Date:   Tue, 4 Sep 2012 12:49:10 +0000
>>>>> From:   Hubert Ritzdorf <Hubert.Ritzdorf at emea.nec.com>
>>>>> Reply-To:       Comments on the MPI Standard
>>>>> <mpi-comments at lists.mpi-forum.org>
>>>>> To:     mpi-comments at lists.mpi-forum.org
>>>>> <mpi-comments at lists.mpi-forum.org>
>>>>>
>>>>>
>>>>>
>>>>> Hi,
>>>>>
>>>>> is it correct that MPI_Win_free should free the memory allocated in
>>>>> MPI_Win_allocate_shared ?
>>>>>
>>>>> MPI_Win_free (Page 416, Line 5-6) specifies this only for
>>>>> memory allocated by MPI_Win_allocate.
>>>>> I think that an additional reference to MPI_Win_allocate_shared is
>>>>> missing also on Page 420, Line 31, Page 448 Line 25, Page 609 Line 31,
>>>>> Page 613 Line 1, Page 614 Line 9, Page 614 Line 31.
>>>>>
>>>>> Hubert
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> mpi3-rma mailing list
>>>>> mpi3-rma at lists.mpi-forum.org
>>>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>>>>
>>>>
>>>>
>>>>
>>> _______________________________________________
>>> mpi3-rma mailing list
>>> mpi3-rma at lists.mpi-forum.org
>>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>>
>>
>>
>> --
>> Jeff Hammond
>> Argonne Leadership Computing Facility
>> University of Chicago Computation Institute
>> jhammond at alcf.anl.gov / (630) 252-5381
>> http://www.linkedin.com/in/jeffhammond
>> https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond
>> _______________________________________________
>> mpi3-rma mailing list
>> mpi3-rma at lists.mpi-forum.org
>> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
>
>
> _______________________________________________
> mpi3-rma mailing list
> mpi3-rma at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma



-- 
Jeff Hammond
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
http://www.linkedin.com/in/jeffhammond
https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond