Fortran cross-reference correction (ticket 0 / errata)

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2014-08-12 10:24:16 -0500

Description

The list of mentioned problems was incomplete. Level is ticket-0.

History

Correction of wrong references about Fortran and split collective I/O.

This ticket should be handled by the language-binding working group and I/O chapter committee.

Extended Scope

None. (No need to add these changes to the erratas' document.)

Proposed Solution

-MPI-3.0 Section 13.4.5, page 524 lines 24-28 read*

Split collective routines must specify a buffer in both the begin and end routines. By specifying the buffer that receives data in the end routine, we can avoid the problems described in "A Problem with Code Movements and Register Optimization," Section 17.1.17 on page 632, but not all of the problems described in Section 17.1.16 on page 631.

-but should read*

Split collective routines must specify a buffer in both the begin and end routines. By specifying the buffer that receives data in the end routine, we can avoid the problems described in "A Problem with Code Movements and Register Optimization," Section 17.1.17 on page 632, but not all of the problems such as those described in Sections 17.1.12, 17.1.13, and 17.1.16 on pages 626, 629, and 631.
Alternative Solutions

None.

Impact on Implementations

None required.

Impact on Applications / Users

None.

Entry for the Change Log

None.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2014-08-13 02:49:49 -0500

This Fortran ticket contains a/some part/s of #427 that was split off into tickets #440 - #446.

It should be accepted by the Fortran/Language Binding and I/O chapter committee / WG before the Sep. 2014 meeting in Japan or voted there with a single errata vote.

mpiforumbot commented 8 years ago

Originally by schulzm on 2014-09-02 03:01:56 -0500

I think this should be just ticket 0 and doesn't have to be voted on.

mpiforumbot commented 8 years ago

Originally by jsquyres on 2014-12-05 09:27:01 -0600

I agree:

As currently stated, this is ticket 0 and doesn't need voting.
Missing a comma after "17.1.13" (this is also ticket 0).
If we're going to cross-reference sections, we need to include them all. That being said, we continually talk about not wanting to cross-reference all sections. So how about changing the phraseology to be a little softer, from:

but not all of the problems described in Sections 17.1.12, 17.1.13 and 17.1.16 on pages 626, 629, and 631.

to

but not all of the problems __such as those__ described in Sections 17.1.12, 17.1.13 and 17.1.16 on pages 626, 629, and 631.

(which I think is still a ticket 0-level change)

mpiforumbot commented 8 years ago

Originally by jsquyres on 2014-12-09 16:35:26 -0600

Fortran WG: we agree this is ticket 0 (including comment 4). It is up to Rolf to convince the IO chapter committee that this is ticket 0.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2014-12-09 18:06:10 -0600

Jeff's corrections are now part of the ticket's solution.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2014-12-16 05:08:12 -0600

During the MPI Forum Meeting Dec. 2014, I sent out the following email to the I/O chapter committee (they were not present at the meeting):

Dear Quincey, Mohamad, Rajeev, and Dries,

I ask you as official I/O chapter to vote positive to the ticket-0 errata in ticket #440 https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/440

The Fortran WG decided that this ticket-0 correction is needed. Split collectives have a special role because of the double using of the buffer argument on both, ..._begin and ..._end routines.

Please can you send a short email reply that I hopefully can pass the ticket at the end.

Best regards Rolf

I received the following answers during the meeting:

Rajeev: OK with me.
Mohamad: Looks good to me.
Dries: Same for me.

Quincey did not answered.

With the okay from the Fortran WG and the three positive answers from the I/O chapter committee, I set this ticket-0 errata to status "passed".

mpiforumbot commented 8 years ago

Originally by jsquyres on 2015-02-04 09:05:27 -0600

I note that the page numbers have already been removed from the MPI-3.1 document in this specific paragraph -- as of r1921, it says:

By specifying the buffer that receives data in the end routine, we can avoid the problems described in “A Problem with Code Movements and Register Optimization,” Section 17.1.17, but not all of the problems described in Section 17.1.16.

Which I think is fine. Since this is ticket 0, I'm just going to add the "such as those" phrase, meaning that 17.1.6 is not all-inclusive of the problems that can occur.

mpiforumbot commented 8 years ago

Originally by jsquyres on 2015-02-04 09:09:40 -0600

Attachment added: mpi31-report-ticket-440-r1923.pdf (2594.3 KiB) Ticket 0 change added; PDF as of SVN r1923

mpiforumbot commented 8 years ago

Originally by jsquyres on 2015-02-04 09:11:56 -0600

PDF with the ticket 0 attached.

Quincey / Rajeev / Mohamad / Dries -- you need to review this; it's your chapter.

mpiforumbot commented 8 years ago

Originally by chaarawi on 2015-02-04 09:25:57 -0600

Replying to jsquyres:

PDF with the ticket 0 attached.

Quincey / Rajeev / Mohamad / Dries -- you need to review this; it's your chapter.

I thought we did already..

mpiforumbot commented 8 years ago

Originally by jsquyres on 2015-02-04 09:42:01 -0600

Rolf polled you on the idea; I just committed the change to your chapter and posted a PDF showing the change to this ticket. That's what you need to review. :-)

mpiforumbot commented 8 years ago

Originally by chaarawi on 2015-02-04 10:17:22 -0600

Looks good to me. Glad I reviewed again since there was a typo that someone (?) accidentally introduced that messed up the sections in the I/O chapter on starting on page 505..

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2015-02-12 09:05:39 -0600

I just detected that in svn r1923, the new text "s 17.1.12, 17.1.13, and" is still missing.

Therefore back to Passed.

mpiforumbot commented 8 years ago

Originally by chaarawi on 2015-02-12 09:08:34 -0600

Replying to RolfRabenseifner:

I just detected that in svn r1923, the new text "s 17.1.12, 17.1.13, and" is still missing.

Therefore back to Passed.

Rolf, Look at Jeff's reply 8 days ago:

" I note that the page numbers have already been removed from the MPI-3.1 document in this specific paragraph -- as of r1921, it says:

By specifying the buffer that receives data in the end routine, we can avoid the problems described in “A Problem with Code Movements and Register Optimization,” Section 17.1.17, but not all of the problems described in Section 17.1.16.

Which I think is fine. Since this is ticket 0, I'm just going to add the "such as those" phrase, meaning that 17.1.6 is not all-inclusive of the problems that can occur. "

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2015-02-12 09:15:17 -0600

Replying to chaarawi:

Replying to RolfRabenseifner:

I just detected that in svn r1923, the new text "s 17.1.12, 17.1.13, and" is still missing.

Therefore back to Passed.

Rolf, Look at Jeff's reply 8 days ago:

" I note that the page numbers have already been removed from the MPI-3.1 document in this specific paragraph -- as of r1921, it says:

By specifying the buffer that receives data in the end routine, we can avoid the problems described in “A Problem with Code Movements and Register Optimization,” Section 17.1.17, but not all of the problems described in Section 17.1.16.

Which I think is fine. Since this is ticket 0, I'm just going to add the "such as those" phrase, meaning that 17.1.6 is not all-inclusive of the problems that can occur. "

The three section numbers 17.1.12, 17.1.13, and 17.1.16 are the important ones and the ticket was therefore done as it is. Sure that the pagenumbers are removed.

"such as those" is already there in r1923.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2015-02-22 02:22:42 -0600

Corrected in svn #r2010. Ticket should be now completly done in svn. Therefore priority changed to "waiting for pdf reviews".

mpiforumbot commented 8 years ago

Originally by jsquyres on 2015-02-22 05:09:09 -0600

Rolf --

The chapter author already said that they didn't want this change.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2015-02-22 13:45:09 -0600

There was no negative reply to my answer. A not voted-in change to a passed ticket would have required consensus. The history of the ticket is, that "such as those" was additinally added, because we were not 100% sure that the list of the important sections is complete or whether there may be some additional minor aspect in one of the other sections. The words "such as those" was not included to remove the major section numbers.

mpiforumbot commented 8 years ago

Originally by schulzm on 2015-02-25 13:05:45 -0600

In preparation for the meeting next week, I was reading up on this, but I can't quite follow. Which is the final text you are proposing for the chapter? Seems like the "such as those" is the preferred version by the chapter committee, but that's not the one on drafts page. I'll list this as an open issue for the meeting to make sure we don't miss it.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2015-02-26 03:49:26 -0600

Attachment added: mpi31-report-ticket-440-r1923+r2010.pdf (4332.7 KiB) Updated pdf for review.

mpiforumbot commented 8 years ago

Originally by RolfRabenseifner on 2015-02-26 03:50:15 -0600

The final text is the text on the ticket above, but without page numbers, i.e.,

Split collective routines must specify a buffer in both the begin and end routines. By specifying the buffer that receives data in the end routine, we can avoid the problems described in "A Problem with Code Movements and Register Optimization," Section 17.1.17, but not all of the problems such as those described in Sections 17.1.12, 17.1.13, and 17.1.16.

This is identical with the text on current svn r2021, mpi-report.pdf (produced with "make cleandoc"), page 528, line 30-34:

Split collective routines must specify a buffer in both the begin and end routines. By specifying the buffer that receives data in the end routine, we can avoid the problems described in "A Problem with Code Movements and Register Optimization," Section 17.1.17, but not all of the problems, such as those described in Sections 17.1.12, 17.1.13, and 17.1.16.

The only difference, i.e., the additional comma, may be wrong.

The attached mpi31-report-ticket-440-r1923.pdf is obsolete and substituted by mpi31-report-ticket-440-r1923+r2010.pdf .

mpi-forum / mpi-forum-historic