MPI window attribute types

[mpiforumbot]

Originally by jdinan on 2011-07-05 12:22:34 -0500

---

In Section 11.2.5 of the ticket #270 text, pointer types are needed for the C in/output arguments to MPI_Win_get/set_attr for all window attributes. This is a clarification to the MPI-2 standard, not an artifact of the MPI-3 RMA proposal.

Original (ticket #270) text:

In C, calls to MPI_Win_get_attr(win, MPI_WIN_BASE, &base, &ag),
MPI_Win_get_attr(win, MPI_WIN_SIZE, &size, &ag),
MPI_Win_get_attr(win, MPI_WIN_DISP_UNIT, &disp_unit, &ag), 
MPI_Win_get_attr(win, MPI_WIN_CREATE_FLAVOR, &create_kind, &ag), and 
MPI_Win_get_attr(win, MPI_WIN_MODEL, &memory_model, &ag) will return in
base a pointer to the start of the window win, and will return in size,
disp_unit, create_kind, and memory_model pointers to the size,
displacement unit of the window, the kind of routine used to create the
window, and the memory model, respectively. And similarly, in C++
(binding deprecated, see Section 15.2).

The following text should be appended to the paragraph above, page 9, line 41:

A detailed listing of the type of the pointer in the attribute value
argument to MPI_Win_get_attr and MPI_Win_set_attr is shown in 
Table~\ref{table:c-attr-types}.

The following table and caption (refernced in the above paragraph) should also be added:

||# Attribute ||# C Type ||
|| MPI_WIN_BASE || void * ||
|| MPI_WIN_SIZE || MPI_Aint * ||
|| MPI_WIN_DISP_UNIT || int * ||
|| MPI_WIN_CREATE_FLAVOR || int * ||
|| MPI_WIN_MODEL || int * ||

Caption: C types of attribute value argument to MPI_Win_get_attr
and MPI_Win_set_attr.

[mpiforumbot]

Originally by gropp on 2011-12-13 10:48:24 -0600

---

Do any current MPI implementations do something different from this? These values are the "natural" ones there's a good chance that this is established practice.

[mpiforumbot]

Originally by jdinan on 2011-12-13 11:22:44 -0600

---

Each of these values is defined somewhere in the standard, but it's challenging to locate them all. That said, I wouldn't be surprised if an application or implementation has gotten it wrong. For example, the 2.2 standard, page 509, line 12 defines how MPI_WIN_BASE is handled:

"... in Fortran, MPI_WIN_GET_ATTR(win, MPI_WIN_BASE, val, ag, ierror) will return in val the base address of the window, converted to an integer. In C, MPI_Win_get_attr(win, MPI_WIN_BASE, &p, &ag) will return in p a pointer to the window base, cast to (void *)."

[mpiforumbot]

Originally by jdinan on 2012-01-12 11:32:34 -0600

---

This should have been read at the January meeting in California. Scheduling it for the next meeting.

Replying to jdinan:

[mpiforumbot]

Originally by jdinan on 2012-03-05 14:44:16 -0600

---

This ticket has been withdrawn as a separate proposal and will be incorporated into MPI 3.0 by the chapter committee.

[mpiforumbot]

Originally by jdinan on 2012-07-03 11:13:59 -0500

---

Reopening this as a "passed" (ticket 0) ticket so we don't lose it.

[mpiforumbot]

Originally by htor on 2012-07-03 23:02:06 -0500

---

This ticket is a bit unclear. What are the proposed changes and how shall they be implemented?

[mpiforumbot]

Originally by jdinan on 2012-07-05 09:28:19 -0500

---

I updated the comments to make the new changes a little more clear. This ticket adds a sentence to the window attributes paragraph and the corresponding table, which clarifies the actual type of the void* argument to get/set attr.

This would have been an MPI 2.2 errata, but I was able to find all of these types scattered throughout the standard. So, we decided to handle this as a ticket 0 clarification.

[mpiforumbot]

Originally by htor on 2012-07-05 12:44:17 -0500

---

This is also committed in approved/one-sided now. Please check!

Torsten

[mpiforumbot]

Originally by hubertritzdorf on 2012-07-15 09:32:47 -0500

---

Reviewed in mpi-report-r1244.pdf

[mpiforumbot]

Originally by potluri on 2012-07-16 11:21:44 -0500

---

reviewed in mpi-report-r1300.pdf

[mpiforumbot]

Originally by jsquyres on 2012-07-18 10:42:36 -0500

---

Appears to be fully committed. Moving to "Waiting for PDF reviews".

[mpiforumbot]

Originally by jdinan on 2012-09-24 04:18:44 -0500

---

This change was included in MPI 3.0.

[mpiforumbot]

Originally by RolfRabenseifner on 2013-01-07 11:41:48 -0600

---

Since Sep. 21, 2012, this ticket is included in MPI-3.0 and the pdf is checked according to https://svn.mpi-forum.org/svn/mpi-forum-docs/trunk/meetings/2012-07-jul/mpi3-tickets.xlsx

Therefore, I set the priority to "Ticket complete".