June 2012 - mpi-forum - mailman3.cels.anl.gov

Re: [Mpi-forum] C++ types inaccessible after #281
by Douglas Miller 26 Jun '12

26 Jun '12

Sure, that sounds pretty reasonable. I just wanted to ask whether it seemed strange to have "std::complex" in there. Seems like (slightly) more than "ticket-0" to me, but should be a simple proposal to write. _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Jed Brown <jedbrown(a)mcs.anl .gov> To Sent by: Main MPI Forum mailing list mpi-forum-bounces <mpi-forum(a)lists.mpi-forum.org>, @lists.mpi-forum. cc org Subject Re: [Mpi-forum] C++ types 06/26/2012 03:08 inaccessible after #281 PM Please respond to Main MPI Forum mailing list <mpi-forum@lists. mpi-forum.org> On Tue, Jun 26, 2012 at 10:43 AM, Douglas Miller <dougmill(a)us.ibm.com> wrote: I think the issue is just that how can a standard that does not specify how C++ fits into things (after ticket 281) then go on to define a data type in terms of C++ types? Is there an informal expectation that implementations will continue to provide an mpicxx wrapper? I see no reason for anyone to be attached to std::complex (actually, I think it's very frequently a bad choice), but having *some* way to use complex types with a predefined (because one-sided cripples non-predefined) MPI_Op still seems important. If there is a need for a C (not C++) complex datatype, that should be a new proposal. But that datatype should not, in my opinion, be defined in terms of something like std::complex. If a platform does not support something like C99 complex types, then it will have to implement complex types and ops itself, or be incomplete. As mentioned earlier, implementations making MPI_C_COMPLEX available independent of C99 would be sufficient. Otherwise we are stuck defining our own types for complex, and then can't use one-sided. Some discussion of work-arounds: http://trac.mcs.anl.gov/projects/mpich2/ticket/1525 _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Dave Goodell 26 Jun '12

26 Jun '12

On Jun 26, 2012, at 11:21 AM MDT, Jeff Hammond wrote: > as far as i'm concerned, these types are only valid with the C bindings when a C++ compiler is used. does that solve it? This is basically the way that we handle the situation in the standard where a particular type isn't supported by a C or Fortran compiler (MPI_INTEGER16, MPI_UINT64_T). -Dave > jeff > > On Tue, Jun 26, 2012 at 11:41 AM, Douglas Miller <dougmill(a)us.ibm.com> wrote: >> I've got to raise a concern here. If we start defining MPI datatypes to be C++ types then how will an implementation support that with only a C compiler? I thought the point of 281 was to eradicate C++ from the standard (and start over later if C++ was to be defined in the standard).

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Jeff Hammond 26 Jun '12

26 Jun '12

There is a huge difference between deprecating support for C++ _bindings_ and deprecating support for C++ _altogether_. ï¿½The ticket is about the C++ bindings. ï¿½What Jed and I are trying to do is make sure that the C _bindings_ allow users to work effectively within C++, which means support std::complex<float/double/long double>. ï¿½Such support already exists in MPI 2.2, but it requires changing "::" to "_" so that one can use them with the C interface. It is absolutely true that the Forum wants better C++ bindings and will kill the original ones to make this happen. ï¿½However, you are the first and only person who has suggested that the goal is to remove support for C++ completely. ï¿½I believe that your statement "I thought the point of 281 was to eradicate C++ from the standard" is completely wrong and made with malice aforethought. You're trying to rewrite history to further your campaign to punish C++ programmers. Many members of the Forum have made it clear that the goal is to support C++ via the C bindings. That does not mean eradicating C++ from the standard. I assume that there will be text expressing the Forum's believe that the C bindings are sufficient for C++, so clearly there should be at least one mention of C++ in the standard. As I said before, the right way to support C++ datatypes for the C bindings is to restrict them to when a C++ compiler is used. ï¿½That solves your issue with not having a C++ compiler. I don't see why you're arguing about the scenario when you don't have a C++ compiler. Do you really have this use case as an implementer or are you just making up silly arguments as part of your campaign of hate against C++? It is my understanding that your actions in regards to ticket #281 are approved by IBM. If this ticket is not supported by IBM, then it would make me happy to hear this. Jeff On Tue, Jun 26, 2012 at 2:21 PM, Douglas Miller <dougmill(a)us.ibm.com> wrote: > > Jeff, This is inappropriate. Especially targeting IBM like that, and on a public mailing list. Ticket 281 has had plenty of discussions and passed votes. This is/was a Forum decision. The fact that I agreed with it, and volunteered to do a lot of the work, is why I am voicing my opinion on this. ï¿½We (the Forum) have said all along that anyone that has a better design for C++ bindings should bring that forward and make a proposal. Getting rid of the current, often considered "broken", C++ bindings is a good way to ensure any future proposal is free to change the design. > > > > > _______________________________________________ > Douglas Miller ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½BlueGene Messaging Development > IBM Corp., Rochester, MN USA ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ Bldg 030-2 A401 > dougmill(a)us.ibm.com ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ Douglas Miller/Rochester/IBM > > Jeff Hammond ---06/26/2012 02:09:59 PM---Jeff Hammond <jhammond(a)alcf.anl.gov> > > Jeff Hammond <jhammond(a)alcf.anl.gov> > Sent by: mpi-forum-bounces(a)lists.mpi-forum.org > > 06/26/2012 01:57 PM > > Please respond to > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org> > > > To > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org>, > > > cc > > > Subject > > Re: [Mpi-forum] C++ types inaccessible after #281 > > > Why do you want to make it so damn hard for anyone to use MPI in C++ codes? ï¿½You don't even have to do anything to support C++ because BG-MPI is based upon MPICH2. > > Can't we just for once try to take user needs into consideration instead of trying to reduce the workload of implementers as much as possible? > > Users matter. ï¿½In IBM land, they are called customers. ï¿½I assure you that some of your largest customers care a great deal about using MPI within C++. > > You're going out of your way to try and punish your customers for using C++. ï¿½Can't you just leave them alone? ï¿½Maybe we - the customers - need to leave IBM alone. > > Jeff > > On Tue, Jun 26, 2012 at 1:43 PM, Douglas Miller <dougmill(a)us.ibm.com> wrote: > > I think the issue is just that how can a standard that does not specify how C++ fits into things (after ticket 281) then go on to define a data type in terms of C++ types? If there is a need for a C (not C++) complex datatype, that should be a new proposal. But that datatype should not, in my opinion, be defined in terms of something like std::complex. If a platform does not support something like C99 complex types, then it will have to implement complex types and ops itself, or be incomplete. > > > > > _______________________________________________ > Douglas Miller ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½BlueGene Messaging Development > IBM Corp., Rochester, MN USA ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ Bldg 030-2 A401 > dougmill(a)us.ibm.xn--com1212-roac09bd4637gea ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ Douglas Miller/Rochester/IBM > > Jeff Hammond ---06/26/2012 12:31:52 PM---Jeff Hammond <jhammond(a)alcf.anl.gov> > > Jeff Hammond <jhammond(a)alcf.anl.gov> > Sent by: mpi-forum-bounces(a)lists.mpi-forum.org > > 06/26/2012 12:21 PM > > > > Please respond to > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org> > To > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org>, > > cc > Subject > > Re: [Mpi-forum] C++ types inaccessible after #281 > > > as far as i'm concerned, these types are only valid with the C bindings when a C++ compiler is used. ï¿½does that solve it? > > jeff > > On Tue, Jun 26, 2012 at 11:41 AM, Douglas Miller <dougmill(a)us.ibm.com> wrote: > > I've got to raise a concern here. If we start defining MPI datatypes to be C++ types then how will an implementation support that with only a C compiler? I thought the point of 281 was to eradicate C++ from the standard (and start over later if C++ was to be defined in the standard). > > > _______________________________________________ > Douglas Miller ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½BlueGene Messaging Development > IBM Corp., Rochester, MN USA ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ Bldg 030-2 A401 > dougmill(a)us.ibm.xn--com1212-roac09bd4637gea ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ ï¿½ Douglas Miller/Rochester/IBM > > Jed Brown ---06/26/2012 11:26:16 AM---Jed Brown <jedbrown(a)mcs.anl.gov> > > Jed Brown <jedbrown(a)mcs.anl.gov> > Sent by: mpi-forum-bounces(a)lists.mpi-forum.org > > 06/26/2012 11:14 AM > > > > > > Please respond to > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org> > To > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org>, > > cc > Subject > > Re: [Mpi-forum] C++ types inaccessible after #281 > > > On Tue, Jun 26, 2012 at 6:13 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: > > There are a few C++-specific datatypes for which it would be worth introducing new C datatypes (e.g., MPI_CXX_COMPLEX). ï¿½It's too bad that no one realized this until now, because this won't make the MPI-3 deadline. ï¿½But it could be part of MPI-3.1 or errata, or some such. > > > Two people in offline discussions have suggested that C bindings to the std::complex types (I'm not aware of anything else that #281 will lose) might qualify as ticket 0._______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > > > -- > Jeff Hammond > Argonne Leadership Computing Facility > University of Chicago Computation Institute > jhammond(a)alcf.anl.xn--gov12/-yma79al02v (630) 252-5381 > http://www.linkedin.com/in/jeffhammond > https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond(See attached file: pic32424.gif)_______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > > > -- > Jeff Hammond > Argonne Leadership Computing Facility > University of Chicago Computation Institute > jhammond(a)alcf.anl.xn--gov12/-yma79al02v (630) 252-5381 > http://www.linkedin.com/in/jeffhammond > https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond(See attached file: pic08490.gif)_______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Jed Brown 26 Jun '12

26 Jun '12

On Tue, Jun 26, 2012 at 10:43 AM, Douglas Miller <dougmill(a)us.ibm.com>wrote: > I think the issue is just that how can a standard that does not specify > how C++ fits into things (after ticket 281) then go on to define a data > type in terms of C++ types? Is there an informal expectation that implementations will continue to provide an mpicxx wrapper? I see no reason for anyone to be attached to std::complex (actually, I think it's very frequently a bad choice), but having *some* way to use complex types with a predefined (because one-sided cripples non-predefined) MPI_Op still seems important. > If there is a need for a C (not C++) complex datatype, that should be a > new proposal. But that datatype should not, in my opinion, be defined in > terms of something like std::complex. If a platform does not support > something like C99 complex types, then it will have to implement complex > types and ops itself, or be incomplete. As mentioned earlier, implementations making MPI_C_COMPLEX available independent of C99 would be sufficient. Otherwise we are stuck defining our own types for complex, and then can't use one-sided. Some discussion of work-arounds: http://trac.mcs.anl.gov/projects/mpich2/ticket/1525

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Douglas Miller 26 Jun '12

26 Jun '12

Jeff, This is inappropriate. Especially targeting IBM like that, and on a public mailing list. Ticket 281 has had plenty of discussions and passed votes. This is/was a Forum decision. The fact that I agreed with it, and volunteered to do a lot of the work, is why I am voicing my opinion on this. We (the Forum) have said all along that anyone that has a better design for C++ bindings should bring that forward and make a proposal. Getting rid of the current, often considered "broken", C++ bindings is a good way to ensure any future proposal is free to change the design. _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Jeff Hammond <jhammond(a)alcf.an l.gov> To Sent by: Main MPI Forum mailing list mpi-forum-bounces <mpi-forum(a)lists.mpi-forum.org>, @lists.mpi-forum. cc org Subject Re: [Mpi-forum] C++ types 06/26/2012 01:57 inaccessible after #281 PM Please respond to Main MPI Forum mailing list <mpi-forum@lists. mpi-forum.org> Why do you want to make it so damn hard for anyone to use MPI in C++ codes? You don't even have to do anything to support C++ because BG-MPI is based upon MPICH2. Can't we just for once try to take user needs into consideration instead of trying to reduce the workload of implementers as much as possible? Users matter. In IBM land, they are called customers. I assure you that some of your largest customers care a great deal about using MPI within C+ +. You're going out of your way to try and punish your customers for using C+ +. Can't you just leave them alone? Maybe we - the customers - need to leave IBM alone. Jeff On Tue, Jun 26, 2012 at 1:43 PM, Douglas Miller <dougmill(a)us.ibm.com> wrote: I think the issue is just that how can a standard that does not specify how C++ fits into things (after ticket 281) then go on to define a data type in terms of C++ types? If there is a need for a C (not C++) complex datatype, that should be a new proposal. But that datatype should not, in my opinion, be defined in terms of something like std::complex. If a platform does not support something like C99 complex types, then it will have to implement complex types and ops itself, or be incomplete. _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Inactive hide details for Jeff Hammond ---06/26/2012 12:31:52 PM---Jeff Hammond <jhammond(a)alcf.anl.gov>Jeff Hammond ---06/26/2012 12:31:52 PM---Jeff Hammond <jhammond(a)alcf.anl.gov> Jeff Hammond < jhammond(a)alcf.anl.gov> Sent by: mpi-forum-bounces@lists. mpi-forum.org To Main MPI Forum 06/26/2012 12:21 PM mailing list < mpi-forum(a)lists.mpi- forum.org>, cc Please respond to Main MPI Forum mailing list < Subject mpi-forum(a)lists.mpi-forum.org> Re: [Mpi-forum] C++ types inaccessible after #281 as far as i'm concerned, these types are only valid with the C bindings when a C++ compiler is used. does that solve it? jeff On Tue, Jun 26, 2012 at 11:41 AM, Douglas Miller <dougmill(a)us.ibm.com> wrote: I've got to raise a concern here. If we start defining MPI datatypes to be C++ types then how will an implementation support that with only a C compiler? I thought the point of 281 was to eradicate C++ from the standard (and start over later if C++ was to be defined in the standard). _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Inactive hide details for Jed Brown ---06/26/2012 11:26:16 AM---Jed Brown <jedbrown(a)mcs.anl.gov>Jed Brown ---06/26/2012 11:26:16 AM---Jed Brown <jedbrown(a)mcs.anl.gov> Jed Brown < jedbrown(a)mcs.anl.gov> Sent by: mpi-forum-bounces@lists. To mpi-forum.org Main MPI Forum mailing list < 06/26/2012 11:14 AM mpi-forum(a)lists.mpi- forum.org>, cc Subject Please respond to Re: [Mpi-forum] C++ Main MPI Forum mailing list < types inaccessible mpi-forum(a)lists.mpi-forum.org> after #281 On Tue, Jun 26, 2012 at 6:13 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: There are a few C++-specific datatypes for which it would be worth introducing new C datatypes (e.g., MPI_CXX_COMPLEX). It's too bad that no one realized this until now, because this won't make the MPI-3 deadline. But it could be part of MPI-3.1 or errata, or some such. Two people in offline discussions have suggested that C bindings to the std::complex types (I'm not aware of anything else that #281 will lose) might qualify as ticket 0. _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond(See attached file: pic32424.gif)_______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond(See attached file: pic08490.gif)_______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Jeff Hammond 26 Jun '12

26 Jun '12

Why do you want to make it so damn hard for anyone to use MPI in C++ codes? You don't even have to do anything to support C++ because BG-MPI is based upon MPICH2. Can't we just for once try to take user needs into consideration instead of trying to reduce the workload of implementers as much as possible? Users matter. In IBM land, they are called customers. I assure you that some of your largest customers care a great deal about using MPI within C++. You're going out of your way to try and punish your customers for using C++. Can't you just leave them alone? Maybe we - the customers - need to leave IBM alone. Jeff On Tue, Jun 26, 2012 at 1:43 PM, Douglas Miller <dougmill(a)us.ibm.com> wrote: > I think the issue is just that how can a standard that does not specify > how C++ fits into things (after ticket 281) then go on to define a data > type in terms of C++ types? If there is a need for a C (not C++) complex > datatype, that should be a new proposal. But that datatype should not, in > my opinion, be defined in terms of something like std::complex. If a > platform does not support something like C99 complex types, then it will > have to implement complex types and ops itself, or be incomplete. > > > > _______________________________________________ > Douglas Miller BlueGene Messaging Development > IBM Corp., Rochester, MN USA Bldg 030-2 A401 > dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM > > [image: Inactive hide details for Jeff Hammond ---06/26/2012 12:31:52 > PM---Jeff Hammond <jhammond(a)alcf.anl.gov>]Jeff Hammond ---06/26/2012 > 12:31:52 PM---Jeff Hammond <jhammond(a)alcf.anl.gov> > > > *Jeff Hammond <jhammond(a)alcf.anl.gov>* > Sent by: mpi-forum-bounces(a)lists.mpi-forum.org > > 06/26/2012 12:21 PM > Please respond to > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org> > > > > To > > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org>, > > > cc > > > Subject > > > Re: [Mpi-forum] C++ types inaccessible after #281 > > > as far as i'm concerned, these types are only valid with the C bindings > when a C++ compiler is used. does that solve it? > > jeff > > On Tue, Jun 26, 2012 at 11:41 AM, Douglas Miller <*dougmill(a)us.ibm.com*<dougmill(a)us.ibm.com>> > wrote: > > I've got to raise a concern here. If we start defining MPI datatypes > to be C++ types then how will an implementation support that with only a C > compiler? I thought the point of 281 was to eradicate C++ from the standard > (and start over later if C++ was to be defined in the standard). > > > _______________________________________________ > Douglas Miller BlueGene Messaging Development > IBM Corp., Rochester, MN USA Bldg 030-2 A401* > **dougmill(a)us.ibm.com* <dougmill(a)us.ibm.com> Douglas > Miller/Rochester/IBM > > [image: Inactive hide details for Jed Brown ---06/26/2012 11:26:16 > AM---Jed Brown <jedbrown(a)mcs.anl.gov>]Jed Brown ---06/26/2012 11:26:16 > AM---Jed Brown <*jedbrown(a)mcs.anl.gov* <jedbrown(a)mcs.anl.gov>> > > > > *Jed Brown <**jedbrown(a)mcs.anl.gov* <jedbrown(a)mcs.anl.gov>*>* > Sent by: *mpi-forum-bounces(a)lists.mpi-forum.org*<mpi-forum-bounces(a)lists.mpi-forum.org> > > > 06/26/2012 11:14 AM > > > > Please respond to > > Main MPI Forum mailing list <*mpi-forum(a)lists.mpi-forum.org*<mpi-forum(a)lists.mpi-forum.org> > > > To > > Main MPI Forum mailing list <*mpi-forum(a)lists.mpi-forum.org*<mpi-forum(a)lists.mpi-forum.org>>, > > > cc > Subject > > > Re: [Mpi-forum] C++ types inaccessible after #281 > > > > On Tue, Jun 26, 2012 at 6:13 AM, Jeff Squyres <*jsquyres(a)cisco.com*<jsquyres(a)cisco.com>> > wrote: > There are a few C++-specific datatypes for which it would be worth > introducing new C datatypes (e.g., MPI_CXX_COMPLEX). It's too bad that no > one realized this until now, because this won't make the MPI-3 deadline. > But it could be part of MPI-3.1 or errata, or some such. > > Two people in offline discussions have suggested that C bindings to > the std::complex types (I'm not aware of anything else that #281 will lose) > might qualify as ticket 0. > _______________________________________________ > mpi-forum mailing list* > **mpi-forum(a)lists.mpi-forum.org* <mpi-forum(a)lists.mpi-forum.org>* > **http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum*<http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum> > > _______________________________________________ > mpi-forum mailing list* > **mpi-forum(a)lists.mpi-forum.org* <mpi-forum(a)lists.mpi-forum.org>* > **http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum*<http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum> > > > > > -- > Jeff Hammond > Argonne Leadership Computing Facility > University of Chicago Computation Institute* > **jhammond(a)alcf.anl.gov* <jhammond(a)alcf.anl.gov> / (630) 252-5381* > **http://www.linkedin.com/in/jeffhammond*<http://www.linkedin.com/in/jeffhammond> > * > * > *https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond*<https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond> > *(See attached file: pic32424.gif)* > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Douglas Miller 26 Jun '12

26 Jun '12

I think the issue is just that how can a standard that does not specify how C++ fits into things (after ticket 281) then go on to define a data type in terms of C++ types? If there is a need for a C (not C++) complex datatype, that should be a new proposal. But that datatype should not, in my opinion, be defined in terms of something like std::complex. If a platform does not support something like C99 complex types, then it will have to implement complex types and ops itself, or be incomplete. _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Jeff Hammond <jhammond(a)alcf.an l.gov> To Sent by: Main MPI Forum mailing list mpi-forum-bounces <mpi-forum(a)lists.mpi-forum.org>, @lists.mpi-forum. cc org Subject Re: [Mpi-forum] C++ types 06/26/2012 12:21 inaccessible after #281 PM Please respond to Main MPI Forum mailing list <mpi-forum@lists. mpi-forum.org> as far as i'm concerned, these types are only valid with the C bindings when a C++ compiler is used. does that solve it? jeff On Tue, Jun 26, 2012 at 11:41 AM, Douglas Miller <dougmill(a)us.ibm.com> wrote: I've got to raise a concern here. If we start defining MPI datatypes to be C++ types then how will an implementation support that with only a C compiler? I thought the point of 281 was to eradicate C++ from the standard (and start over later if C++ was to be defined in the standard). _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Inactive hide details for Jed Brown ---06/26/2012 11:26:16 AM---Jed Brown <jedbrown(a)mcs.anl.gov>Jed Brown ---06/26/2012 11:26:16 AM---Jed Brown < jedbrown(a)mcs.anl.gov> Jed Brown < jedbrown(a)mcs.anl.gov> Sent by: mpi-forum-bounces@lists. mpi-forum.org To Main MPI Forum 06/26/2012 11:14 AM mailing list < mpi-forum(a)lists.mpi- forum.org>, cc Please respond to Subject Main MPI Forum mailing list < mpi-forum(a)lists.mpi-forum.org> Re: [Mpi-forum] C++ types inaccessible after #281 On Tue, Jun 26, 2012 at 6:13 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: There are a few C++-specific datatypes for which it would be worth introducing new C datatypes (e.g., MPI_CXX_COMPLEX). It's too bad that no one realized this until now, because this won't make the MPI-3 deadline. But it could be part of MPI-3.1 or errata, or some such. Two people in offline discussions have suggested that C bindings to the std::complex types (I'm not aware of anything else that #281 will lose) might qualify as ticket 0._______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum _______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond(See attached file: pic32424.gif)_______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Jeff Hammond 26 Jun '12

26 Jun '12

as far as i'm concerned, these types are only valid with the C bindings when a C++ compiler is used. does that solve it? jeff On Tue, Jun 26, 2012 at 11:41 AM, Douglas Miller <dougmill(a)us.ibm.com>wrote: > I've got to raise a concern here. If we start defining MPI datatypes to be > C++ types then how will an implementation support that with only a C > compiler? I thought the point of 281 was to eradicate C++ from the standard > (and start over later if C++ was to be defined in the standard). > > > _______________________________________________ > Douglas Miller BlueGene Messaging Development > IBM Corp., Rochester, MN USA Bldg 030-2 A401 > dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM > > [image: Inactive hide details for Jed Brown ---06/26/2012 11:26:16 > AM---Jed Brown <jedbrown(a)mcs.anl.gov>]Jed Brown ---06/26/2012 11:26:16 > AM---Jed Brown <jedbrown(a)mcs.anl.gov> > > > *Jed Brown <jedbrown(a)mcs.anl.gov>* > Sent by: mpi-forum-bounces(a)lists.mpi-forum.org > > 06/26/2012 11:14 AM > Please respond to > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org> > > > > To > > > Main MPI Forum mailing list <mpi-forum(a)lists.mpi-forum.org>, > > > cc > > > Subject > > > Re: [Mpi-forum] C++ types inaccessible after #281 > > > On Tue, Jun 26, 2012 at 6:13 AM, Jeff Squyres <*jsquyres(a)cisco.com*<jsquyres(a)cisco.com>> > wrote: > > There are a few C++-specific datatypes for which it would be worth > introducing new C datatypes (e.g., MPI_CXX_COMPLEX). It's too bad that no > one realized this until now, because this won't make the MPI-3 deadline. > But it could be part of MPI-3.1 or errata, or some such. > > > Two people in offline discussions have suggested that C bindings to the > std::complex types (I'm not aware of anything else that #281 will lose) > might qualify as ticket 0._______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > > > _______________________________________________ > mpi-forum mailing list > mpi-forum(a)lists.mpi-forum.org > http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum > -- Jeff Hammond Argonne Leadership Computing Facility University of Chicago Computation Institute jhammond(a)alcf.anl.gov / (630) 252-5381 http://www.linkedin.com/in/jeffhammond https://wiki.alcf.anl.gov/parts/index.php/User:Jhammond

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Jed Brown 26 Jun '12

26 Jun '12

On Tue, Jun 26, 2012 at 8:41 AM, Douglas Miller <dougmill(a)us.ibm.com> wrote: > I've got to raise a concern here. If we start defining MPI datatypes to be > C++ types then how will an implementation support that with only a C > compiler? I thought the point of 281 was to eradicate C++ from the standard > (and start over later if C++ was to be defined in the standard). The C++ standard specifies the data representation so I don't see any reason an implementation couldn't opt to implement the predefined MPI_Ops without actually using std::complex. An alternative approach is to note that C99 _Complex has the same data representation as std::complex, then ask implementations to provide MPI_C_COMPLEX even on systems without C99 (making it usable from C++). This isn't really a complicated thing, but we definitely have users of complex types on Windows and cannot use one-sided operations unless a predefined complex type is available there.

1 0

Re: [Mpi-forum] C++ types inaccessible after #281
by Douglas Miller 26 Jun '12

26 Jun '12

I've got to raise a concern here. If we start defining MPI datatypes to be C++ types then how will an implementation support that with only a C compiler? I thought the point of 281 was to eradicate C++ from the standard (and start over later if C++ was to be defined in the standard). _______________________________________________ Douglas Miller BlueGene Messaging Development IBM Corp., Rochester, MN USA Bldg 030-2 A401 dougmill(a)us.ibm.com Douglas Miller/Rochester/IBM Jed Brown <jedbrown(a)mcs.anl .gov> To Sent by: Main MPI Forum mailing list mpi-forum-bounces <mpi-forum(a)lists.mpi-forum.org>, @lists.mpi-forum. cc org Subject Re: [Mpi-forum] C++ types 06/26/2012 11:14 inaccessible after #281 AM Please respond to Main MPI Forum mailing list <mpi-forum@lists. mpi-forum.org> On Tue, Jun 26, 2012 at 6:13 AM, Jeff Squyres <jsquyres(a)cisco.com> wrote: There are a few C++-specific datatypes for which it would be worth introducing new C datatypes (e.g., MPI_CXX_COMPLEX). It's too bad that no one realized this until now, because this won't make the MPI-3 deadline. But it could be part of MPI-3.1 or errata, or some such. Two people in offline discussions have suggested that C bindings to the std::complex types (I'm not aware of anything else that #281 will lose) might qualify as ticket 0._______________________________________________ mpi-forum mailing list mpi-forum(a)lists.mpi-forum.org http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi-forum

1 0