Subject: Re: MPI-2.1 ballot #2 To: mpi-21@XXXXXXXXXXXXX X-Mailer: Lotus Notes Release 5.0.5 September 22, 2000 From: "Richard Treumann" Date: Fri, 8 Mar 2002 10:41:38 -0500 X-MIMETrack: Serialize by Router on D01ML064/01/M/IBM(Release 5.0.9a |January 28, 2002) at 03/08/2002 10:43:14 AM MIME-Version: 1.0 Content-type: text/plain; charset=us-ascii Sender: owner-mpi-21@XXXXXXXXXXXXX Precedence: bulk Reply-To: mpi-21@XXXXXXXXXXXXX I am voting yes but do think there is a small remaining question regarding MPI_Put etc to MPI_PROC_NULL. When using FENCE synchronization, a FENCE that ends an epoch may or may not begin another fenced epoch. The resolution is based on the RMA operations done after it. I.E. MPI_Win_fence; MPI_Put; MPI_Win_fence; MPI_Put; MPI_Win_start is an error because the second MPI_Put has implicitly begun a fenced epoch and an MPI_Win_start cannot be done until the epoch which the second MPI_Put belongs to is fenced. Does an MPI_Put/Get/Accumulate to MPI_PROC_NULL have the same side effect? I think it must or we have made things more complex, not less. The analogy to point-to-point does not give me a clear answer. Dick Dick Treumann RS/6000 SP Development IBM Poughkeepsie Unix Development Lab Dept 0lva / MS P963 -- 2455 South Road -- Poughkeepsie, NY 12601 Tele (845) 433-7846 Fax (845) 433-8363 Rusty Lusk cc: lusk@XXXXXXXXXXX Sent by: Subject: MPI-2.1 ballot #2 owner-mpi-21@mpi- forum.org 03/05/2002 05:29 PM Please respond to mpi-21 Please sent us your votes by Friday, March 8. X I vote in favor of the changes and additions suggested below ___ I vote against the changes and additions suggested below ___ I would prefer a separate vote on item ___ (Please don't do this lightly.) There is already a backlog of new suggestions for clarifications that have been discussed in recent email. We hope to get a ballot for those together shortly. Regards, Rusty Lusk and Bill Gropp ====================================================================== 6: Page 114, after line 4, add MPI_PROC_NULL is a valid target rank in the MPI RMA calls MPI_ACCUMULATE, MPI_GET, and MPI_PUT. The effect is the same as for MPI_PROC_NULL in MPI point-to-point communication. ========================================================================