Ok thanks very much for helping me clarify the issue.
Best regards, Marina On Wed, Jul 27, 2016 at 9:21 PM, Benjamin Herta <bhe...@us.ibm.com> wrote: > Different MPI implementations have different capabilities. Some can handle > multiple threads making MPI calls in parallel (MPI_THREAD_MULTIPLE), some > require only one thread accessing MPI at a time (MPI_THREAD_SERIALIZED), > and some only work if all calls to MPI go through the same thread > (MPI_THREAD_FUNNELED: other threads can exist, but should never make MPI > calls), or if there is only one thread in the program (MPI_THREAD_SINGLE). > > Our PAMI backend can support the FUNNELED case, by using a special thread > which handles all probing, and other threads can hand off message sending > operations to that same thread. But we don't currently support this same > capability in the MPI backend, since it's such a rare case and more > difficult to implement. We do support the other three modes. > > See https://www.open-mpi.org/doc/v2.0/man3/MPI_Init_thread.3.php for more > details on the different modes. > > > Since your system only supports MPI_THREAD_SINGLE, you can only have one > thread per place, which as you noted earlier, requires the combination of > X10_STATIC_THREADS=1 X10_NTHREADS=1 and X10_NUM_IMMEDIATE_THREADS=0. > Unfortunately, as you also noted, this means you can't use clocks in your > program. > > I believe the only path forward here is to re-work your application so > that it doesn't use clocks. > > > - Ben > > [image: Inactive hide details for Marina Andric ---07/27/2016 03:05:14 > PM---Hi Josh, I checked and the level provided is not serialized]Marina > Andric ---07/27/2016 03:05:14 PM---Hi Josh, I checked and the level > provided is not serialized, as I thought (sorry), > > From: Marina Andric <marina.and...@imtlucca.it> > To: Mailing list for users of the X10 programming language < > x10-users@lists.sourceforge.net> > Date: 07/27/2016 03:05 PM > > Subject: Re: [X10-users] Problem with compiling programs to C++ backend > after building x10 from source > ------------------------------ > > > > Hi Josh, > > I checked and the level provided is not serialized, as I thought (sorry), > it is MPI_THREAD_SINGLE. > > Is it compatible with X10 implementation? > > Thanks a lot, > Marina > > > On Wed, Jul 27, 2016 at 7:20 PM, Joshua J Milthorpe <*jjmil...@us.ibm.com* > <jjmil...@us.ibm.com>> wrote: > > Hi Marina, > > it looks like you are setting the environment variable correctly. I > guess it's possible that your MPI implementation doesn't support > MPI_THREAD_SERIALIZED, either. What version of MPI are you using? (and what > sort of cluster are you running on?) Can you try a simple MPI program that > initializes MPI with MPI_THREAD_SERIALIZED, to see if this works outside of > X10? > > *https://www.mpi-forum.org/docs/mpi-3.1/mpi31-report/node303.htm* > <https://www.mpi-forum.org/docs/mpi-3.1/mpi31-report/node303.htm> > > Cheers, > > Josh > > > [image: Inactive hide details for Marina Andric ---07/27/2016 12:53:35 > PM---Hi Josh, Thanks a lot for your email.]Marina Andric ---07/27/2016 > 12:53:35 PM---Hi Josh, Thanks a lot for your email. > > From: Marina Andric <*marina.and...@imtlucca.it* > <marina.and...@imtlucca.it>> > To: Mailing list for users of the X10 programming language < > *x10-users@lists.sourceforge.net* <x10-users@lists.sourceforge.net>> > Date: 07/27/2016 12:53 PM > > > Subject: Re: [X10-users] Problem with compiling programs to C++ > backend after building x10 from source > ------------------------------ > > > > > Hi Josh, > > Thanks a lot for your email. > > I just tested it again and faced the old mistake again: > > > * "Underlying MPI implementation does not provide requested threading > level Unable to support requested level of X10 > threading; exiting"* > > > I am struggling to understand why this happens. I opened x10rt_mpi.cc > file and saw the part of the code that generates the error, which is > when level_required != level_provided, where level_required is set > according to what getenv(X10RT_MPI_THREAD_SERIALIZED) returns. > > Am I maybe setting the variable in a wrong way? I am executing a > script with a line > > export X10RT_MPI_THREAD_SERIALIZED=1 > > Thanks, > Marina > > > > On Wed, Jul 27, 2016 at 6:29 PM, Joshua J Milthorpe < > *jjmil...@us.ibm.com* <jjmil...@us.ibm.com>> wrote: > Hi Marina, > > the message you found regarding X10RT_MPI_THREAD_MULTIPLE is now > out of date :-( > > Since X10 2.5.2, MPI_THREAD_MULTIPLE is the default threading > level, but it is possible to request MPI_THREAD_SERIALIZED by > setting the > environment variable X10RT_MPI_THREAD_SERIALIZED. In this case, X10 > will > serialize all calls to MPI functions through a global lock (in > accordance > with the rules for MPI_THREAD_SERIALIZED). > > *https://xtenlang.atlassian.net/browse/XTENLANG-3490* > <https://xtenlang.atlassian.net/browse/XTENLANG-3490> > > Could you please try running with X10RT_MPI_THREAD_SERIALIZED=1 > and let us know the result? > > Cheers, > *Josh Milthorpe* > Post Doctoral Researcher > Cognitive Systems: Learning to Reason > IBM Research > > > ------------------------------ > *Phone:* *1-914-945-2209* <1-914-945-2209> > *E-mail:* *jjmil...@us.ibm.com* <jjmil...@us.ibm.com> > 1101 Kitchawan Rd > Yorktown Heights, NY 10598 > United States > > > > [image: Inactive hide details for Marina Andric ---07/27/2016 > 12:13:16 PM---Hi again, I came across]Marina Andric > ---07/27/2016 12:13:16 PM---Hi again, I came across > > From: Marina Andric <*marina.and...@imtlucca.it* > <marina.and...@imtlucca.it>> > To: Mailing list for users of the X10 programming language < > *x10-users@lists.sourceforge.net* > <x10-users@lists.sourceforge.net>> > Date: 07/27/2016 12:13 PM > Subject: Re: [X10-users] Problem with compiling programs to C++ > backend after building x10 from source > ------------------------------ > > > > > Hi again, > > I came across > > "Only if X10_STATIC_THREADS=1 X10_NTHREADS=1 and > X10_NUM_IMMEDIATE_THREADS=0 will we use single-threaded MPI modes." > > on one of the pages dedicated to X10. > > I now initialized the variables and I'm not anymore facing the > error that provided and requested threading levels do not match. > However, my code uses Clocks for synchronization and I get > runtime error > > *ClockUseException: Clocks are not > compatible with static threads. * > > I guess that I've hit the wall now and can't proceed in this > direction. > > Or is there a possible workaround to make Clocks work in mpi > thread serialized setting ? > > Thank you, > Marina > > > > > > > > > > On Wed, Jul 27, 2016 at 3:26 PM, Marina Andric < > *marina.and...@imtlucca.it* <marina.and...@imtlucca.it>> wrote: > Hi, > > Thanks a lot Marco for the help. Yes, I initially > copied the executable, then I built it on the cluster > and now it works, > almost :) > > Works well until the moment I compile the program > with *-x10rt mpi* option. The program is compiled > and executable is generated, but it is not running. The > error states: > > > * "Underlying MPI implementation does not > provide requested threading level > Unable to > support requested level of X10 threading; exiting"* > > (the whole output is attached) > > I found this page: > *https://sourceforge.net/p/x10/mailman/message/33490796/* > <https://sourceforge.net/p/x10/mailman/message/33490796/> > and started looking into X10 runtime variables > (X10RT_MPI_THREAD_MULTIPLE, > X10_STATIC_THREADS) and played a bit with those but with > no success.. > > On my linux laptop it works with -x10rt mpi option, > with no errors. In both cases, on my machine and the > cluster, > MPI_THREAD_MULTIPLE is set to no. > > Does someone has an idea how to solve this? > > Thank you, > Marina > > > > > > > > > > > > > > On Tue, Jul 26, 2016 at 1:00 AM, Marco Bungart < > *m.bung...@gmx.net* <m.bung...@gmx.net>> wrote: > Hi Maina, > > looks like you have not built the program on the > cluster. Did you just > copy the executable? This will in general not work. > You should download > and compile/install X10 on the cluster as well. The > easiest way would be > to build X10 from source (tutorial found here: > *http://x10-lang.org/x10-development/building-x10-from-source.html* > > <http://x10-lang.org/x10-development/building-x10-from-source.html> > ). > After that, re-compile your program on the cluster > and execute it. > > Cheers, > Marco > > Am 25.07.2016 um 18:11 schrieb Marina Andric: > > > > Thanks a lot Marco, this was something I neglected > :) the program is > > compiling/executing now on my machine. > > > > 'Im still having a problem on the cluster. I get > compilation error stating: > > > > x10.dist/stdlib/lib/libx10.so: undefined reference > to 'memcpy@GLIBC_2.14' > > x10.dist/stdlib/lib/libx10.so: undefined reference > to > > 'clock_gettime@GLIBC_2.17' > > > > Does somebody know the possible origins of this > problem? Do I maybe need > > to load some additional modules on the cluster, > except gcc and mpi? > > > > Warm regards, > > Marina > > > > > > > > > > > > > > > > > > > > > > > > On Sun, Jul 24, 2016 at 11:48 PM, Marco Bungart < > *m.bung...@gmx.net* <m.bung...@gmx.net> > > <mailto:*m.bung...@gmx.net* <m.bung...@gmx.net>>> > wrote: > > > > Hi, > > > > have you tried running "$> ./a.out" instead of > "$> a.out"? Some systems > > distinguish between system-wide executable > (i.e. executable located in > > one of the paths specified in the $PATH > environment variable) and > > executables in the current folder (i.e. "."). > > > > Cheers, > > Marco > > > > Am 24.07.2016 um 21:22 schrieb Marina Andric: > > > Hi all, > > > > > > I am trying to run an x10 program on a > cluster. The cluster supports > > > only MPI based communication between nodes > so I built the latest x10 > > > source (version 2.6.0) with option > -DX10RT_MPI=true and the build was > > > successful. > > > > > > My program works when it's compiled to java > (via x10c) but it doesn't > > > work when it is compiled to c++ (via x10c++). > > > > > > For example, when > > > > > >> x10.dist/bin/x10c++ hello.x10 > (creates a.out) > > >> a.out > > > > > > the outcome is a.out: command not found. > > > > > > I tested a bit around, with -o option, > -x10rt sockets (mpi) and > > nothing > > > seemed to change. I eventually built x10 > without -DX10RT_MPI=true and > > > the problem is the same, also with the x10 > version 2.5.4. > > > > > > I was building and testing on my local linux > machine and I checked all > > > the program versions and they seem to be ok: > > > > > > ant 1.9.3 > > > g++ 4.8.4 > > > > > > Can someone please say if I'm missing > something? > > > > > > Thank you, > > > Marina > > > > > > > > > > > > > > > > > > > > > > > > > > > > ------------------------------------------------------------------------------ > > > What NetFlow Analyzer can do for you? > Monitors network bandwidth > > and traffic > > > patterns at an interface-level. Reveals > which users, apps, and > > protocols are > > > consuming the most bandwidth. Provides > multi-vendor support for > > NetFlow, > > > J-Flow, sFlow and other flows. Make informed > decisions using > > capacity planning > > > reports.*http://sdm.link/zohodev2dev* > <http://sdm.link/zohodev2dev> > > > > > > > > > > > > > _______________________________________________ > > > X10-users mailing list > > > *X10-users@lists.sourceforge.net* > <X10-users@lists.sourceforge.net> > > <mailto:*X10-users@lists.sourceforge.net* > <X10-users@lists.sourceforge.net>> > > > > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > > > > > --- > > Diese E-Mail wurde von Avast > Antivirus-Software auf Viren geprüft. > > *https://www.avast.com/antivirus* > <https://www.avast.com/antivirus> > > > > > > > > ------------------------------------------------------------------------------ > > What NetFlow Analyzer can do for you? Monitors > network bandwidth and > > traffic > > patterns at an interface-level. Reveals which > users, apps, and > > protocols are > > consuming the most bandwidth. Provides > multi-vendor support for NetFlow, > > J-Flow, sFlow and other flows. Make informed > decisions using > > capacity planning > > reports.*http://sdm.link/zohodev2dev* > <http://sdm.link/zohodev2dev> > > _______________________________________________ > > X10-users mailing list > > *X10-users@lists.sourceforge.net* > <X10-users@lists.sourceforge.net> <mailto: > *X10-users@lists.sourceforge.net* > <X10-users@lists.sourceforge.net>> > > > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > > > > > > > > > > ------------------------------------------------------------------------------ > > What NetFlow Analyzer can do for you? Monitors > network bandwidth and traffic > > patterns at an interface-level. Reveals which > users, apps, and protocols are > > consuming the most bandwidth. Provides > multi-vendor support for NetFlow, > > J-Flow, sFlow and other flows. Make informed > decisions using capacity planning > > reports.*http://sdm.link/zohodev2dev* > <http://sdm.link/zohodev2dev> > > > > > > > > _______________________________________________ > > X10-users mailing list > > *X10-users@lists.sourceforge.net* > <X10-users@lists.sourceforge.net> > > > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > > --- > Diese E-Mail wurde von Avast Antivirus-Software auf > Viren geprüft. > *https://www.avast.com/antivirus* <https://www.avast.com/antivirus> > > > > > ------------------------------------------------------------------------------ > What NetFlow Analyzer can do for you? Monitors > network bandwidth and traffic > patterns at an interface-level. Reveals which users, > apps, and protocols are > consuming the most bandwidth. Provides multi-vendor > support for NetFlow, > J-Flow, sFlow and other flows. Make informed > decisions using capacity planning > reports.*http://sdm.link/zohodev2dev* > <http://sdm.link/zohodev2dev> > _______________________________________________ > X10-users mailing list > *X10-users@lists.sourceforge.net* <X10-users@lists.sourceforge.net> > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > ------------------------------------------------------------------------------ > What NetFlow Analyzer can do for you? Monitors network bandwidth > and traffic > patterns at an interface-level. Reveals which users, apps, and > protocols are > consuming the most bandwidth. Provides multi-vendor support for > NetFlow, > J-Flow, sFlow and other flows. Make informed decisions using > capacity planning > reports. > > *http://sdm.link/zohodev2dev_______________________________________________* > > <http://sdm.link/zohodev2dev_______________________________________________> > X10-users mailing list > *X10-users@lists.sourceforge.net* <X10-users@lists.sourceforge.net> > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > > > > ------------------------------------------------------------------------------ > What NetFlow Analyzer can do for you? Monitors network bandwidth > and traffic > patterns at an interface-level. Reveals which users, apps, and > protocols are > consuming the most bandwidth. Provides multi-vendor support for > NetFlow, > J-Flow, sFlow and other flows. Make informed decisions using > capacity planning > reports.*http://sdm.link/zohodev2dev* > <http://sdm.link/zohodev2dev> > _______________________________________________ > X10-users mailing list > *X10-users@lists.sourceforge.net* <X10-users@lists.sourceforge.net> > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > > > ------------------------------------------------------------------------------ > What NetFlow Analyzer can do for you? Monitors network bandwidth and > traffic > patterns at an interface-level. Reveals which users, apps, and > protocols are > consuming the most bandwidth. Provides multi-vendor support for > NetFlow, > J-Flow, sFlow and other flows. Make informed decisions using capacity > planning > reports. > > *http://sdm.link/zohodev2dev_______________________________________________* > > <http://sdm.link/zohodev2dev_______________________________________________> > X10-users mailing list > *X10-users@lists.sourceforge.net* <X10-users@lists.sourceforge.net> > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > > > > ------------------------------------------------------------------------------ > What NetFlow Analyzer can do for you? Monitors network bandwidth and > traffic > patterns at an interface-level. Reveals which users, apps, and > protocols are > consuming the most bandwidth. Provides multi-vendor support for > NetFlow, > J-Flow, sFlow and other flows. Make informed decisions using capacity > planning > reports.*http://sdm.link/zohodev2dev* <http://sdm.link/zohodev2dev> > _______________________________________________ > X10-users mailing list > *X10-users@lists.sourceforge.net* <X10-users@lists.sourceforge.net> > *https://lists.sourceforge.net/lists/listinfo/x10-users* > <https://lists.sourceforge.net/lists/listinfo/x10-users> > > > ------------------------------------------------------------------------------ > _______________________________________________ > X10-users mailing list > X10-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/x10-users > > > >
------------------------------------------------------------------------------
_______________________________________________ X10-users mailing list X10-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/x10-users
