October 2013 - mpi-forum - mailman3.cels.anl.gov

MPI_Pack and large counts
by Rob Latham 17 Oct '13

17 Oct '13

The definition for MPI_Pack looks like this: int MPI_Pack(const void* inbuf, int incount, MPI_Datatype datatype, void *outbuf, int outsize, int *position, MPI_Comm comm) What happens if incount and datatype describe something larger than an int? Looks to me like 'outsize' and 'posistion' will overflow, but I'm hoping that I'm just missing something. Likewise, MPI_Unpack: int MPI_Unpack(const void* inbuf, int insize, int *position, void *outbuf, int outcount, MPI_Datatype datatype, MPI_Comm comm) ==rob -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA

1 0

Re: [Mpi-forum] Clarification on MPI communicator semantics
by Scott Pakin 11 Oct '13

11 Oct '13

On 10/11/2013 07:00 PM, Pavan Balaji wrote: > It is valid to do such a comparison. This is mentioned on page 12, lines 38-39. > > "MPI procedures that operate on opaque objects are passed handle arguments to access these objects. In addition to their use by MPI calls for object access, handles can participate in assignments and comparisons." Ah, I missed that (obviously). However, I should point out that that text didn't have any of the keywords I was searching for -- except "communicator", which appears too often in the document to be a meaningful search term. Do you therefore think it might make sense for the MPI_Comm_compare() description to refer back to that section? Perhaps it could distinguish an "==" that returns true from an MPI_Comm_compare() that returns MPI_IDENT. (I'm sure there's some way in MPI to produce two different handles to the same communicator object, but I can't think of what that might be. MPI_Comm_f2c() perhaps?) Thanks for the quick reponse! -- Scott

1 0

Re: [Mpi-forum] Clarification on MPI communicator semantics
by Pavan Balaji 11 Oct '13

11 Oct '13

It is valid to do such a comparison. This is mentioned on page 12, lines 38-39. "MPI procedures that operate on opaque objects are passed handle arguments to access these objects. In addition to their use by MPI calls for object access, handles can participate in assignments and comparisons." -- Pavan On Oct 11, 2013, at 7:56 PM, Scott Pakin wrote: > Is it valid or invalid to compare MPI communicators with "==" (in C)? > > I didn't see anything in the MPI 3.0 standard that addressed that > question. I know I can use MPI_Comm_compare() and check for > MPI_IDENT, but "==" sure is a lot more convenient. If the answer is > that an "==" comparison of two MPI_Comm types isn't guaranteed to > compile or to return correct results, then how should I compare a > communicator to MPI_COMM_NULL, which the Argonne MPI page claims is > not a valid argument to MPI_Comm_compare() (cf. > http://www.mcs.anl.gov/research/projects/mpi/www/www3/MPI_Comm_compare.html… > > Thanks, > -- Scott > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Pavan Balaji http://www.mcs.anl.gov/~balaji

1 0

Clarification on MPI communicator semantics
by Scott Pakin 11 Oct '13

11 Oct '13

Is it valid or invalid to compare MPI communicators with "==" (in C)? I didn't see anything in the MPI 3.0 standard that addressed that question. I know I can use MPI_Comm_compare() and check for MPI_IDENT, but "==" sure is a lot more convenient. If the answer is that an "==" comparison of two MPI_Comm types isn't guaranteed to compile or to return correct results, then how should I compare a communicator to MPI_COMM_NULL, which the Argonne MPI page claims is not a valid argument to MPI_Comm_compare() (cf. http://www.mcs.anl.gov/research/projects/mpi/www/www3/MPI_Comm_compare.html… Thanks, -- Scott

1 0

MPI Forum list is back
by Jeff Squyres (jsquyres) 10 Oct '13

10 Oct '13

DNS has been restored, so once it propagates out to you: - mail to the Forum mailing lists is flowing again - the meetings and lists web sites are back The main www.mpi-forum.org mailing list is still offline, but they're working on it. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0

Unplanned mpi-forum.org DNS/web/mail outage
by Jeff Squyres (jsquyres) 10 Oct '13

10 Oct '13

Several of you have asked me "why is www.mpi-forum.org down?", and "why is all @lists.mpi-forum.org mail bouncing?". It looks like there is an ongoing attempt to migrate the main Forum web sites (?and DNS?) from Argonne to U. Illinois. I'm not sure if something went wrong, or if this is just part of the migration plan. I asked Argonne/Illinois about it yesterday when these problems started appearing, but haven't yet received an answer. Indiana U., who maintains the Forum mailing lists, has *temporarily* made the main mpi-forum(a)lists.mpi-forum.org mailing list available at mpi-forum(a)open-mpi.org so that I can send out this status message. When the "lists.mpi-forum.org" DNS name is properly restored, Indiana U. will move the mailing list back. So be aware that all MPI Forum traffic -- with the exception of this one list -- is temporarily down. Hopefully Argonne/Illinois can resolve the situation soon. WG leads: upon request, we can have IU send you a list of all the email addresses on your @lists.mpi-forum.org listserv. -- Jeff Squyres jsquyres(a)cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/

1 0