- discuss - mailman3.cels.anl.gov

MPI_WIN_LOCK and MPI_MODE_NOCHECK assert
by Jaroslaw Zola 20 Feb '13

20 Feb '13

I have a question regarding implementation of the MPI_MODE_NOCHECK assertion for MPI_WIN_LOCK. Is current mpich implementation taking advantage of this hint? Also, according to the standard: MPI_MODE_NOCHECK - no other process holds, or will attempt to acquire, a conflicting lock, while the caller holds the window lock. This is useful when mutual exclusion is achieved by other means, but the coherence operations that may be attached to the lock and unlock calls are still required. Are two concurrent locks from two different MPI processes followed by MPI_GET (i.e. given window is not modified but read only) considered conflicting? Thanks! - Jaric

1 0

Re: [mpich-discuss] Fwd: A MPICH beginner's question,
by Jayesh Krishna 19 Feb '13

19 Feb '13

Hi, If you have "Win7 64bit" please use "mpich2-1.4.1p1-win-x86-64.msi" (Uninstall all other versions of MPICH2 from your machine). If you installed MPICH2 successfully you shouldn't have to run SMPD manually (See the installer's guide on how to install MPICH2 via an administrator command prompt). When compiling your code please specify the path to include/lib paths correctly in the VS environment. The 64-bit version of MPICH2 would get installed in "c:\Program Files\MPICH2\" directory by default. Regards, Jayesh ----- Original Message ----- From: "Peng Zhang" <spkevabn(a)gmail.com> To: mpich-discuss(a)mcs.anl.gov Sent: Monday, February 18, 2013 2:56:24 AM Subject: [mpich-discuss] Fwd: A MPICH beginner's question, Jayesh Krishna : Hi ,thank you for reading my E-mail, I'm a Chinese hydrodynamic student and helping my teacher to upgrade an old software, but I am facing a problem now. ：（ My system condition is : Win7 64bit + VS2010 + IVF Composer XE 2011 I downloaded "mpich2-1.4.1p1-win-ia3" and "mpich2-1.4.1p1-win-x86-64" ( I have tried both two .msi) I have run smpd.exe to install MPICH (run as Administrator) and registered the servere (smpd.exe -regserver) After creating a new IVF Console Application project, I set : 1 、 Win32 to x64, 2 、 Set "Fortran\General\Additional Include Directories" to "C:\Program Files (x86)\MPICH2\include;" (I have tried 64bit too,but return the same result) 3 、 Set"Linker\General\Additional Library Directories" to "C:\Program Files (x86)\MPICH2\lib;" 4 、 Set"Linker\Input\ Additional Dependencies" to "mpi.lib" Then I wrote a piece code of "Hello World" 。 program Lab include 'mpif.h' character * (MPI_MAX_PROCESSOR_NAME) processor_name integer myid,numprocs,namelen,rc,ierr call MPI_INIT(ierr) call MPI_COMM_RANK(MPI_COMM_WORLD,myid,ierr) call MPI_COMM_SIZE(MPI_COMM_WORLD,numprocs,ierr) call MPI_GET_PROCESSOR_NAME(processor_name,namelen,ierr) write(*,10) myid,numprocs,processor_name 10 FORMAT('Hello World!Process',I2,'of',I1,'on',20A) call MPI_FINALIZE(rc) end program Lab But there is a error "Cannot open mpi.lib" 内嵌图片 1 I'm sure that the file path is correct (I have tried 64bit already.) 内嵌图片 2 I have search the answer in Google for two days, unfortunaly I'm failed. I request you to give me some hints,Please. *__*.. Best regards _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] Get universe size no longer supported?
by Reuti 18 Feb '13

18 Feb '13

Hi Jeff, Am 18.02.2013 um 15:02 schrieb Jeff Hammond: > If you need MPI_UNIVERSE_SIZE to be defined, you can set it explicitly > via "mpiexec -usize MPI_UNIVERSE_SIZE". thanks for the hint. In particular -usize SYSTEM seems to restore the former behavior. -- Reuti > There very well could be a bug in MPICH right now, but there is at > least a solution at hand and the MPI standard is not violated. > > Jeff > > On Mon, Feb 18, 2013 at 6:44 AM, Reuti <reuti(a)staff.uni-marburg.de> wrote: >> Am 18.02.2013 um 13:37 schrieb Jeff Hammond: >> >>> I've informed the appropriate people about the Trac issue. >> >> Thanks. >> >> >>> I recall that MPI stipulates only that MPI_UNIVERSE_SIZE exist; it >>> need not be defined. >>> >>> I think that MPICH might set MPI_UNIVERSE_SIZE in the process manager. >>> What PM are you using? >> >> I started with a plain `mpiexec -np 4 ./program`, hence it should use Hydra as it's a symbolic link to mpiexec.hydra. >> >> -- Reuti >> >> >>> Jeff >>> >>> On Mon, Feb 18, 2013 at 5:52 AM, Reuti <Reuti(a)staff.uni-marburg.de> wrote: >>>> Hi, >>>> >>>> a call: >>>> >>>> error=MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universe_sizep, &flag); >>>> >>>> worked in MPICH2 1.4.1p1. But the "flag" returned unsupported in 1.5 and 3.0.2. Was this info removed from being retrieved? >>>> >>>> -- Reuti >>>> >>>> PS: I found http://trac.mpich.org/projects/mpich/ticket/1426 mentioning MPI_UNIVERSE_SIZE but it throws an error: >>>> >>>> TracError: The Trac Environment needs to be upgraded. >>>> >>>> Run "trac-admin /www/trac.mpich.org/projects/mpich upgrade" >>>> _______________________________________________ >>>> discuss mailing list discuss(a)mpich.org >>>> To manage subscription options or unsubscribe: >>>> https://lists.mpich.org/mailman/listinfo/discuss >>> >>> >>> >>> -- >>> 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 >>> _______________________________________________ >>> discuss mailing list discuss(a)mpich.org >>> To manage subscription options or unsubscribe: >>> https://lists.mpich.org/mailman/listinfo/discuss >> >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss > > > > -- > 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 > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] Get universe size no longer supported?
by Jeff Hammond 18 Feb '13

18 Feb '13

If you need MPI_UNIVERSE_SIZE to be defined, you can set it explicitly via "mpiexec -usize MPI_UNIVERSE_SIZE". There very well could be a bug in MPICH right now, but there is at least a solution at hand and the MPI standard is not violated. Jeff On Mon, Feb 18, 2013 at 6:44 AM, Reuti <reuti(a)staff.uni-marburg.de> wrote: > Am 18.02.2013 um 13:37 schrieb Jeff Hammond: > >> I've informed the appropriate people about the Trac issue. > > Thanks. > > >> I recall that MPI stipulates only that MPI_UNIVERSE_SIZE exist; it >> need not be defined. >> >> I think that MPICH might set MPI_UNIVERSE_SIZE in the process manager. >> What PM are you using? > > I started with a plain `mpiexec -np 4 ./program`, hence it should use Hydra as it's a symbolic link to mpiexec.hydra. > > -- Reuti > > >> Jeff >> >> On Mon, Feb 18, 2013 at 5:52 AM, Reuti <Reuti(a)staff.uni-marburg.de> wrote: >>> Hi, >>> >>> a call: >>> >>> error=MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universe_sizep, &flag); >>> >>> worked in MPICH2 1.4.1p1. But the "flag" returned unsupported in 1.5 and 3.0.2. Was this info removed from being retrieved? >>> >>> -- Reuti >>> >>> PS: I found http://trac.mpich.org/projects/mpich/ticket/1426 mentioning MPI_UNIVERSE_SIZE but it throws an error: >>> >>> TracError: The Trac Environment needs to be upgraded. >>> >>> Run "trac-admin /www/trac.mpich.org/projects/mpich upgrade" >>> _______________________________________________ >>> discuss mailing list discuss(a)mpich.org >>> To manage subscription options or unsubscribe: >>> https://lists.mpich.org/mailman/listinfo/discuss >> >> >> >> -- >> 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 >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- 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: [mpich-discuss] Get universe size no longer supported?
by Reuti 18 Feb '13

18 Feb '13

Am 18.02.2013 um 13:37 schrieb Jeff Hammond: > I've informed the appropriate people about the Trac issue. Thanks. > I recall that MPI stipulates only that MPI_UNIVERSE_SIZE exist; it > need not be defined. > > I think that MPICH might set MPI_UNIVERSE_SIZE in the process manager. > What PM are you using? I started with a plain `mpiexec -np 4 ./program`, hence it should use Hydra as it's a symbolic link to mpiexec.hydra. -- Reuti > Jeff > > On Mon, Feb 18, 2013 at 5:52 AM, Reuti <Reuti(a)staff.uni-marburg.de> wrote: >> Hi, >> >> a call: >> >> error=MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universe_sizep, &flag); >> >> worked in MPICH2 1.4.1p1. But the "flag" returned unsupported in 1.5 and 3.0.2. Was this info removed from being retrieved? >> >> -- Reuti >> >> PS: I found http://trac.mpich.org/projects/mpich/ticket/1426 mentioning MPI_UNIVERSE_SIZE but it throws an error: >> >> TracError: The Trac Environment needs to be upgraded. >> >> Run "trac-admin /www/trac.mpich.org/projects/mpich upgrade" >> _______________________________________________ >> discuss mailing list discuss(a)mpich.org >> To manage subscription options or unsubscribe: >> https://lists.mpich.org/mailman/listinfo/discuss > > > > -- > 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 > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Re: [mpich-discuss] Get universe size no longer supported?
by Jeff Hammond 18 Feb '13

18 Feb '13

I've informed the appropriate people about the Trac issue. I recall that MPI stipulates only that MPI_UNIVERSE_SIZE exist; it need not be defined. I think that MPICH might set MPI_UNIVERSE_SIZE in the process manager. What PM are you using? Jeff On Mon, Feb 18, 2013 at 5:52 AM, Reuti <Reuti(a)staff.uni-marburg.de> wrote: > Hi, > > a call: > > error=MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universe_sizep, &flag); > > worked in MPICH2 1.4.1p1. But the "flag" returned unsupported in 1.5 and 3.0.2. Was this info removed from being retrieved? > > -- Reuti > > PS: I found http://trac.mpich.org/projects/mpich/ticket/1426 mentioning MPI_UNIVERSE_SIZE but it throws an error: > > TracError: The Trac Environment needs to be upgraded. > > Run "trac-admin /www/trac.mpich.org/projects/mpich upgrade" > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss -- 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

Get universe size no longer supported?
by Reuti 18 Feb '13

18 Feb '13

Hi, a call: error=MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_UNIVERSE_SIZE, &universe_sizep, &flag); worked in MPICH2 1.4.1p1. But the "flag" returned unsupported in 1.5 and 3.0.2. Was this info removed from being retrieved? -- Reuti PS: I found http://trac.mpich.org/projects/mpich/ticket/1426 mentioning MPI_UNIVERSE_SIZE but it throws an error: TracError: The Trac Environment needs to be upgraded. Run "trac-admin /www/trac.mpich.org/projects/mpich upgrade"

1 0

Fwd: A MPICH beginner's question,
by Peng Zhang 18 Feb '13

18 Feb '13

Jayesh Krishna <mpich-discuss(a)mcs.anl.gov>: Hi ,thank you for reading my E-mail, I'm a Chinese hydrodynamic student and helping my teacher to upgrade an old software, but I am facing a problem now. ：（ My system condition is : Win7 64bit + VS2010 + IVF Composer XE 2011 I downloaded "mpich2-1.4.1p1-win-ia3" and "mpich2-1.4.1p1-win-x86-64" ( I have tried both two .msi) I have run smpd.exe to install MPICH (run as Administrator) and registered the servere (smpd.exe -regserver) After creating a new IVF Console Application project, I set : 1、Win32 to x64, 2、 Set "Fortran\General\Additional Include Directories" to "C:\Program Files (x86)\MPICH2\include;" (I have tried 64bit too,but return the same result) 3、 Set"Linker\General\Additional Library Directories" to "C:\Program Files (x86)\MPICH2\lib;" 4、Set"Linker\Input\ Additional Dependencies" to "mpi.lib" Then I wrote a piece code of "Hello World" 。 program Lab include 'mpif.h' character * (MPI_MAX_PROCESSOR_NAME) processor_name integer myid,numprocs,namelen,rc,ierr call MPI_INIT(ierr) call MPI_COMM_RANK(MPI_COMM_WORLD,myid,ierr) call MPI_COMM_SIZE(MPI_COMM_WORLD,numprocs,ierr) call MPI_GET_PROCESSOR_NAME(processor_name,namelen,ierr) write(*,10) myid,numprocs,processor_name 10 FORMAT('Hello World!Process',I2,'of',I1,'on',20A) call MPI_FINALIZE(rc) end program Lab But there is a error "Cannot open mpi.lib" [image: 内嵌图片 1] I'm sure that the file path is correct (I have tried 64bit already.) [image: 内嵌图片 2] I have search the answer in Google for two days, unfortunaly I'm failed. I request you to give me some hints,Please. *__*.. Best regards

1 0

Re: [mpich-discuss] mpich-3.0.2 and Windows
by Andrey V. Malyshev 16 Feb '13

16 Feb '13

Hello, Jayesh and All! You wrote to <discuss(a)mpich.org> on Fri, 15 Feb 2013 12:02:28 -0600 (CST): JK> Hi, JK> Due to limited developer time available/devoted to MPICH2 JK> support/development on Windows the build scripts are out-of-date. We JK> currently don't have a time line on when (if) these scripts will get JK> updated. Sorry for the inconvenience. Unfortunately, my skills are not enough to create a patch for Windows scripts, but maybe someone more qualified users could help, relying on tips from the developers? Hey, enthusiasts! :) I can make some tests, if needed. JK> ----- Original Message ----- JK> From: "Andrey V. Malyshev" <amal(a)icm.krasn.ru> JK> To: discuss(a)mpich.org JK> Sent: Friday, February 15, 2013 7:51:55 AM JK> Subject: [mpich-discuss] mpich-3.0.2 and Windows JK> Hello! JK> Is it possible to build actual release (3.0.2) on Windows? JK> Unfortunately current scripts seems to be unusable. JK> Or maybe Windows compatibility has broken intentionally? JK> I have tryed to build and a lot of errors occurs. JK> -- JK> With best regards, Andrey V. Malyshev. JK> E-mail: amal(a)krasn.ru JK> _______________________________________________ JK> discuss mailing list discuss(a)mpich.org JK> To manage subscription options or unsubscribe: JK> https://lists.mpich.org/mailman/listinfo/discuss -- With best regards, Andrey V. Malyshev. E-mail: amal(a)krasn.ru

1 0

Re: [mpich-discuss] valgrind is indicating problems in MPICH2 when using the MPE library
by Dave Goodell 15 Feb '13

15 Feb '13

On Feb 15, 2013, at 2:49 PM CST, John Michael Alex Grime wrote: > If I compile an MPI code using the "-llmpe" and "-lmpe" options to enable MPI tracing, running the resultant code through valgrind ( "mpiexec -np X valgrind --track-origins=yes ./program ..." ) produces several instances of the following type of message: […] > Has anyone else noticed any problems with MPE using uninitialized stack variables? It's probably a bug in MPE, unfortunately. I don't think we have funding to work on MPE right now, though we would welcome patches if someone wants to dive in and fix the issue. I've filed a ticket so that the issue isn't totally lost: https://trac.mpich.org/projects/mpich/ticket/1794 Thanks for letting us know, -Dave

1 0