Ticket #499 (closed bug: fixed)
Add test for ANY_SOURCE + ANY_TAG
| Reported by: | William Gropp <wgropp@…> | Owned by: | gropp |
|---|---|---|---|
| Priority: | major | Milestone: | mpich2-1.1.1 |
| Component: | mpich2 | Keywords: | |
| Cc: |
Description
This is a reminder to add this as a test case in our test suite. Bill Begin forwarded message: > From: Jeremy Enos <jenos@ncsa.uiuc.edu> > Date: March 24, 2009 4:17:37 PM CDT > To: "bp-users@ncsa.uiuc.edu" <bp-users@ncsa.uiuc.edu> > Subject: from [Seid Koric <skoric@ncsa.uiuc.edu>]] > Reply-To: Jeremy Enos <jenos@ncsa.uiuc.edu> > > > All, > > FYI from an IBM researcher on bp. It might be useful for some of us > having problems with mpi buffers since the January upgrade. > > > > >> Seid, this is a bug in MPI! That's why it showed up only after the >> system upgrade. There is a problem in the current MPI; when a recv >> is posted with MPI_ANY_SOURCE and MPI_ANY_TAG, the recv buffer does >> not contain the correct message. >> >> For example: >> >> 0 : mysendi to 8 for msg type 20 of len 12 >> 0 : 1 5 1 0 12 4 507104 0 1522431 3 14 300 >> 0 : wsi_send: ssize, dest, styp = 12 8 20 >> >> 8 : Recd type 20 from 0 >> 8 : 0 0 0 0 0 0 0 0 0 0 0 0 >> >> 0 sent a msg to 8 containing 12 integers: 1 5 1 0 12 4 507104 0 >> 1522431 3 14 300 >> but 8 recd all 0s. This happens only with mpi_recv containing >> MPI_ANY_SOURCE and MPI_ANY_TAG, which is usually rare. >> >> Anyway, I will do a workaround and then we should be able to bet >> nastran-b results, Right now, this bug prevents WSMP from running on >> any matrix. >> >> -anshul >> -- >> Anshul Gupta >> Business Analytics & Mathematical Sciences >> IBM T.J. Watson Research Center Tel: +1-914-945-1450 >> 1101 Kitchawan Road / Route 134 Fax: +1-914-945-3434 >> Yorktown Heights, NY 10598 Email: anshul@us.ibm.com >> >> Inactive hide details for Seid Koric <skoric@ncsa.uiuc.edu&g >> Seid Koric <skoric@ncsa.uiuc.edu> >> >> Seid Koric <skoric@ncsa.uiuc.edu> >> >> 03/23/2009 12:15 PM >> [] >> >> To >> >> Anshul Gupta/Watson/IBM@IBMUS >> [] >> >> cc >> >> [] >> >> Subject >> >> Re: nastran-b >> >> Anhsul, >> >> wsmp with nastran-b could not run, I got mpi errors of this type >> for all runs: >> >> ERROR: 0032-117 User pack or receive buffer is too small (4) in >> MPI_Recv, task 16 >> >> So I did not run slu or mumps with nastran-b either. >> >> Seid >> >> >> > > --=====================_78903031==_ > Content-Type: application/octet-stream; name="4b3f1b7.gif" > Content-ID: <7.0.0.16.2.20090324083439.04eb5738@ncsa.uiuc.edu.3> > Content-Transfer-Encoding: base64 > Content-Disposition: inline; filename="4b3f1b7.gif" > > R0lGODlhEAAQAKECAMzMzAAAAP///wAAACH5BAEAAAIALAAAAAAQABAAAAIXlI+py > +0PopwxUbpu > ZRfKZ2zgSJbmSRYAIf4fT3B0aW1pemVkIGJ5IFVsZWFkIFNtYXJ0U2F2ZXIhAAA7 > --=====================_78903031==_ > Content-Type: application/octet-stream; name="4b3f1c7.gif" > Content-ID: <7.0.0.16.2.20090324083439.04eb5738@ncsa.uiuc.edu.4> > Content-Transfer-Encoding: base64 > Content-Disposition: inline; filename="4b3f1c7.gif" > > R0lGODlhEAABAIAAAAAAAP///yH5BAEAAAEALAAAAAAQAAEAAAIEjI8ZBQA7 > --=====================_78903031==_ > Content-Type: text/plain; charset="us-ascii"; format=flowed > > > --=====================_78903031==_-- > > William Gropp Deputy Director for Research Institute for Advanced Computing Applications and Technologies Paul and Cynthia Saylor Professor of Computer Science University of Illinois Urbana-Champaign
Attachments
Change History
Note: See
TracTickets for help on using
tickets.
