-
Notifications
You must be signed in to change notification settings - Fork 3
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Updated RMA chapter #270
Comments
Originally by htor on 2011-10-26 11:30:21 -0500 Brian's implementation is available at the following SVN repository: http://svn.open-mpi.org/svn/ompi/tmp-public/mpi3-onesided/ See his email to the mpi-forum mailinglist from Wed, 12 Oct 2011 for details! Torsten |
Originally by moody20 on 2011-10-27 19:39:12 -0500 A bunch of ticket 0 changes ... (based on one-side-2.2.pdf) There are several sentences where we use the term "memory registration", but we don't define what that means. The text only talks about "attaching memory", so we should use this term instead, especially since "memory registration" is commonly used in modern machines to mean something similar but different. p7, 40 p7, 42 p7, 44 p8, 2 p8, 5 p9, 29-30 p9, 39 p9, 42 p13, 18 p22, 41-42 p23, 6-7 Question: Would it be useful to set fields for count and get elements? p45, 38 p52, 36 p60, 27 |
Originally by htor on 2011-11-21 17:37:37 -0600 Attachment added: |
Originally by htor on 2011-11-21 17:38:55 -0600 Hi Adam,
|
Originally by jsquyres on 2012-01-10 11:21:18 -0600 Torsten tells me that this implementation is complete. |
Originally by htor on 2012-05-29 01:41:48 -0500 Committed a long time ago, Torsten |
Originally by RolfRabenseifner on 2012-06-29 12:53:40 -0500 appLang committed (SVN 1210) |
Originally by jsquyres on 2012-07-03 09:17:04 -0500 Reviewed applang commit. |
Originally by RolfRabenseifner on 2012-07-10 09:08:44 -0500 Comment https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/284#comment:24 It is fixed together with the #284 and marked with 229.5. |
Originally by hubertritzdorf on 2012-07-15 09:00:36 -0500 Documents mpi-report-r1244.pdf) (and mpi-report-r1270.pdf |
Originally by htor on 2012-07-15 16:38:52 -0500 Replying to hubertritzdorf:
Thanks & Best, |
Originally by hubertritzdorf on 2012-07-16 04:29:28 -0500 The additional paragraph break is inconsistent to the other examples in Section 11.8 and |
Originally by htor on 2012-07-16 09:23:16 -0500 Replying to hubertritzdorf:
http://www.unixer.de/sec/mpi-report-r1301.pdf Torsten |
Originally by potluri on 2012-07-16 10:58:32 -0500 Review based off mpi-report-r1300.pdf. Below are a few comments. Thanks. Page 428: Line 8 - Page 435: Line 13 - Page 437: Line 15 "MPI_COMPARE_AND_SWAP performs a remote compare and swap operation" --> MPI_COMPARE_AND_SWAP performs an remote atomic compare and swap operation Page 449 There is no mention of atomicity in the description after MPI_COMPARE_AND_SWAP. However, atomic compare and swap is mentioned in the text before it. I hope this will suffice but thought of bringing it up Page 475 Lines 3-5 "If fence or post-start-complete-wait synchronization is used, updates to a public window copy can be delayed in both memory models until the window owner executes a synchronization call." this contradicts with the text below from Page 474 Lines 30-31 "In the RMA unied memory model, an update of a location in a private window in process memory becomes visible without additional RMA calls" Sorry if I am taking it out of context Page 475 Line 20 "multiple MPI_PUT operation" --> multiple MPI_PUT operations |
Originally by potluri on 2012-07-16 13:03:16 -0500 One more minor comment based on mpi-report-r1300.pdf Page 452: Line 31 |
Originally by htor on 2012-07-17 11:12:31 -0500 Replying to potluri:
Comment 21 is also applied. Please check http://www.unixer.de/sec/mpi-report.pdf . Thanks, |
Originally by htor on 2012-07-17 12:38:02 -0500 The issue with MPI_Win_flush is fixed now after discussion at the Forum. Please check: http://www.unixer.de/sec/mpi-report.pdf Torsten |
Originally by jdinan on 2012-07-17 18:04:08 -0500 Reviewed r1300, minus the examples. Changes published to mailing list in pdf markup and have been applied. |
Originally by goodell on 2012-07-17 21:33:13 -0500 MPI_Win_flush and friends review: OK as of Torsten's PDF. |
Originally by hubertritzdorf on 2012-07-18 09:17:44 -0500 MPI_Win_flush and related function changes review: OK |
Originally by jsquyres on 2012-07-18 09:58:24 -0500 Appears to be fully committed. Moving to "Waiting for PDF reviews". |
Originally by RolfRabenseifner on 2013-01-07 11:22:33 -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, by proxy / on behalf of the ticket owner, I close it with priority "Ticket complete", resolution "Text committed". |
Originally by gropp on 2011-03-14 08:32:23 -0500
The RMA working group has extensively revised the RMA chapter to add, in a way consistent with the MPI-2 RMA and MPI in general, new RMA support. These include ways to attach memory to an MPI_Win, perform read-modify-write operations, and use requests on RMA operations. In addition, in response to many requests, many operations previously defined as erroneous now have undefined (but not erroneous) behavior, which is similar to other one-sided programming models. As the chapter has been extensively revised, this ticket is for the entire chapter, and has been accepted by the RMA working group (chapter committee).
Minor edits to this chapter are in #275.
"Ticket 0" changes are expected to be submitted separately, though some updates have been made while updating the text of the chapter.
In addition to the new one sided chapter, the following changes are added to Annex A:
Error classes:
Collective operations:
Attributes
Add a new table, "MPI Window Create Flavors":
Add a new table, "MPI Window Models":
Note no C++ values unless C++ is undeprecated.
The change log entry is:
Substantial revision of the entire One-sided chapter, with new routines for window creation, additional synchronization methods in passive target, new one-sided communication routines, a new memory model, and other changes.
The text was updated successfully, but these errors were encountered: