November 2013 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] Problem about debug with idb!
by Wesley Bland 18 Nov '13

18 Nov '13

Unless I'm mistaken, there is no -dbg runtime option. Also, the configure flag you're using is for debugging MPICH itself, not the application, so you can safely ignore it. Debugging parallel applications is always tricky. Short of using a real parallel debugger (Totalview, DDT, etc.), you can use some tricks to get gdb to work. 1) mpiexec -n 1 gdb <application> : -n <nun procs - 1> <application> This will start up the application and attach a debugger to process 0. I think this will only work if rank 0 is local, but that's often enough for debugging purposes. 2) Insert a sleep or while loop in the code of the target process. If you cause the process that you want to debug to busy wait, you can give yourself enough time to ssh to it and attach a debugger with something like 'gdb --pid <pid>'. There are probably other tricks that would work, but these are the two most common. Give them a shot and one should work for you. Thanks, Wesley > On Nov 17, 2013, at 10:55 PM, "ylniu(a)iccas.ac.cn" <ylniu(a)iccas.ac.cn> wrote: > > Dear all, > > I want to debug parallel fortran program with idb. > > The configure parameter of MPICH2 or MPICH3 is as follows: > > ./configure --prefix=$HOME/program/mpich/mpich2-1.4.1p1 \ > --enable-f77 --enable-fc F77=ifort FC=ifort \ > FFLAGS="-g -traceback -checkall" FCFLAGS="-g -traceback -checkall" \ > --enable-static --enable-g=dbg \ > --enable-traceback --enable-f90modules > > or > > ./configure --prefix=$HOME/program/mpich/mpich2-1.4.1p1 \ > --enable-f77 --enable-fc F77=ifort FC=ifort \ > FFLAGS="-g -traceback -checkall" FCFLAGS="-g -traceback -checkall" \ > --enable-static --enable-g=all \ > --enable-traceback --enable-f90modules > > > mpirun -machinefile $PBS_NODEFILE -np $NPROCS -gdb=idb -parallel myprogram -npool 1 > > There comes the error message: > ======================================================================================= > match_arg (./utils/args/args.c:160): unrecognized argument dbg > HYDU_parse_array (./utils/args/args.c:175): argument matching returned error > parse_args (./ui/mpich/utils.c:1609): error parsing input array > HYD_uii_mpx_get_parameters (./ui/mpich/utils.c:1660): unable to parse user arguments > main (./ui/mpich/mpiexec.c:153): error parsing parameters > ======================================================================================= > > How can mpirun recognition the argument "dbg"? > > Best regards! > > ylniu(a)iccas.ac.cn > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss

1 0

Problem about debug with idb!
by ylniu＠iccas.ac.cn 17 Nov '13

17 Nov '13

Dear all, I want to debug parallel fortran program with idb. The configure parameter of MPICH2 or MPICH3 is as follows: ./configure --prefix=$HOME/program/mpich/mpich2-1.4.1p1 \ --enable-f77 --enable-fc F77=ifort FC=ifort \ FFLAGS="-g -traceback -checkall" FCFLAGS="-g -traceback -checkall" \ --enable-static --enable-g=dbg \ --enable-traceback --enable-f90modules or ./configure --prefix=$HOME/program/mpich/mpich2-1.4.1p1 \ --enable-f77 --enable-fc F77=ifort FC=ifort \ FFLAGS="-g -traceback -checkall" FCFLAGS="-g -traceback -checkall" \ --enable-static --enable-g=all \ --enable-traceback --enable-f90modules mpirun -machinefile $PBS_NODEFILE -np $NPROCS -gdb=idb -parallel myprogram -npool 1 There comes the error message: ======================================================================================= match_arg (./utils/args/args.c:160): unrecognized argument dbg HYDU_parse_array (./utils/args/args.c:175): argument matching returned error parse_args (./ui/mpich/utils.c:1609): error parsing input array HYD_uii_mpx_get_parameters (./ui/mpich/utils.c:1660): unable to parse user arguments main (./ui/mpich/mpiexec.c:153): error parsing parameters ======================================================================================= How can mpirun recognition the argument "dbg"? Best regards! ylniu(a)iccas.ac.cn

1 0

Re: [mpich-discuss] Build issue with MPICH 3.1rc1 using KNEM
by Junchao Zhang 16 Nov '13

16 Nov '13

Hello, Jerome, Thanks for reporting the bug. We will fix it in the upcoming MPICH 3.1rc2 --Junchao Zhang On Nov 16, 2013 12:44 AM, "Jerome Vienne" <viennej(a)tacc.utexas.edu> wrote: > Hi, > > While trying to build MPICH 3.1rc1 with KNEM, I had the following error: > *src/mpid/ch3/channels/nemesis/src/mpid_nem_lmt_dma.c(358): error: > identifier "MPIR_CVAR_NEM_LMT_DMA_THRESHOLD" is undefined* > * nodma = !knem_has_dma || data_sz < MPIR_CVAR_NEM_LMT_DMA_THRESHOLD;* > *compilation aborted for > src/mpid/ch3/channels/nemesis/src/mpid_nem_lmt_dma.c (code 2)* > > I am guessing that the issue is coming from the name of the > variable "MPIR_CVAR_NEM_LMT_DMA_THRESHOLD". Looking at the code, I saw that > the name of this variable should be: "MPIR_CVAR_*NEMESIS* > _LMT_DMA_THRESHOLD". > > I did this change in my code and the build was ok. > > Thanks, > Jerome > -- > Jerome Vienne, Ph.D > HPC Software Tools Group > Texas Advanced Computing Center (TACC) > viennej(a)tacc.utexas.edu | Phone: (512) 475-9322 > Office: ROC 1.455B | Fax: (512) 475-9445 > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss >

1 0

Build issue with MPICH 3.1rc1 using KNEM
by Jerome Vienne 16 Nov '13

16 Nov '13

Hi, While trying to build MPICH 3.1rc1 with KNEM, I had the following error: *src/mpid/ch3/channels/nemesis/src/mpid_nem_lmt_dma.c(358): error: identifier "MPIR_CVAR_NEM_LMT_DMA_THRESHOLD" is undefined* * nodma = !knem_has_dma || data_sz < MPIR_CVAR_NEM_LMT_DMA_THRESHOLD;* *compilation aborted for src/mpid/ch3/channels/nemesis/src/mpid_nem_lmt_dma.c (code 2)* I am guessing that the issue is coming from the name of the variable "MPIR_CVAR_NEM_LMT_DMA_THRESHOLD". Looking at the code, I saw that the name of this variable should be: "MPIR_CVAR_*NEMESIS* _LMT_DMA_THRESHOLD". I did this change in my code and the build was ok. Thanks, Jerome -- Jerome Vienne, Ph.D HPC Software Tools Group Texas Advanced Computing Center (TACC) viennej(a)tacc.utexas.edu | Phone: (512) 475-9322 Office: ROC 1.455B | Fax: (512) 475-9445

1 0

Re: [mpich-discuss] [mvapich-discuss] Non-MPI_THREAD-SINGLE mode with enabled MV2 affinity?
by Jonathan Perkins 15 Nov '13

15 Nov '13

Thanks for the suggestion. This wiki is actually maintained by the MPICH group at ANL. I'm including their discuss list in this reply. I believe that the biggest improvement to the wiki would be to expand on the syntax accepted by user defined binding. On Fri, Nov 15, 2013 at 1:01 PM, Thiago Quirino - NOAA Federal <thiago.quirino(a)noaa.gov> wrote: > Thank you so much, Jonathan. > > Is it possible to add this information to the Wiki? Though I've searched the > net for hours, unfortunately I couldn't find any reference to this. The > closest I found regarding disabling internal MVAPICH2 and mpiexec.hydra > affinities in favor of external MPI process pinning using taskset or numactl > was this: > > https://scicomp.jlab.org/docs/node/66 > > Thank you so much again, Jonathan. > Thiago. > > > > > On Fri, Nov 15, 2013 at 10:31 AM, Jonathan Perkins > <perkinjo(a)cse.ohio-state.edu> wrote: >> >> Hello Thiago. I've confirmed that the following syntax will give you >> what you're asking for. You'll basically need to specify user >> binding, replace your commas with pluses, and replace your colons with >> commas. >> >> [perkinjo@sandy1 install]$ ./bin/mpiexec -n 4 -bind-to >> user:0+1+2+3,4+5+6+7,8+9+10+11,12+13+14+15 -env MV2_ENABLE_AFFINITY=0 >> ./cpumask >> 1111000000000000 >> 0000111100000000 >> 0000000011110000 >> 0000000000001111 >> >> On Wed, Nov 13, 2013 at 7:02 PM, Thiago Quirino - NOAA Federal >> <thiago.quirino(a)noaa.gov> wrote: >> > Hi, Jonathan. >> > >> > Using mpiexec.hydra's binding capability, is it possible to assign a CPU >> > range for each MPI task in a node? Suppose I want to spawn 4 MPI tasks >> > per >> > node, where each node has 2 sockets with 8 cores each (total 16 cores). >> > I >> > want task 1 to run on CPU range 0-3, task 2 on CPU range 4-7, task 3 on >> > CPU >> > range 8-11, and task 4 on CPU range 12-15. I used to accomplish this >> > using >> > the MV2_CPU_BINDING variables as follows: >> > >> > export MV2_CPU_MAPPING=0,1,2,3:4,5,6,7:8,9,10,11:12,13,14,15 >> > >> > Can I accomplish the same binding configuration with mpiexec.hydra's >> > binding >> > capability? I only see socket binding options in the Wiki. >> > >> > Thanks again, Jonathan. >> > Thiago. >> > >> > >> > >> > On Tue, Nov 12, 2013 at 1:56 PM, Jonathan Perkins >> > <jonathan.lamar.perkins(a)gmail.com> wrote: >> >> >> >> Hello Thiago. Perhaps you can try an alternative to >> >> MV2_ENABLE_AFFINITY. If you use the hydra process manager >> >> (mpiexec.hydra), you can disable the library affinity and use the >> >> launcher affinity instead. In this case the other threading levels >> >> will be available to you. >> >> >> >> Please see >> >> >> >> https://wiki.mpich.org/mpich/index.php/Using_the_Hydra_Process_Manager#Proc… >> >> for more information on how to use this hydra feature. Also please do >> >> not forget to set MV2_ENABLE_AFFINITY to 0. >> >> >> >> Please let us know if this helps. >> >> >> >> On Fri, Nov 8, 2013 at 6:41 PM, Thiago Quirino - NOAA Federal >> >> <thiago.quirino(a)noaa.gov> wrote: >> >> > Hi, folks. Quick question about MVAPICH2 and affinity support. >> >> > >> >> > Is it possible to invoke MPI_Init_thread with any mode other than >> >> > "MPI_THREAD_SINGLE" and still use "MV2_ENABLE_AFFINITY=1"? In my >> >> > hybrid >> >> > application I mix MPI with raw Pthreads (not OpenMP). I start 4 MPI >> >> > tasks in >> >> > each 16 cores node, where each node has 2 sockets with 8 Sandybridge >> >> > cores >> >> > each. Each of the 4 MPI tasks then spawns 4 pthreads for a total of >> >> > 16 >> >> > pthreads/node, or 1 pthread/core. Within each MPI task, the MPI calls >> >> > are >> >> > serialized among the 4 pthreads, so I can use any MPI_THREAD_* mode, >> >> > but >> >> > I >> >> > don't know which mode will work best. I want to assign each of the 4 >> >> > MPI >> >> > tasks in a node a set of 4 cores using MV2_CPU_MAPPING (e.g. export >> >> > MV2_CPU_MAPPING=0,1,2,3:4,5,6,7:8,9,10,11:12,13,14,15) so that the 4 >> >> > pthreads spawned by each MPI task can migrate to any processor within >> >> > its >> >> > exclusive CPU set of size 4. >> >> > >> >> > Is that possible with modes other than MPI_THREAD_SINGLE? If not, do >> >> > you >> >> > foresee any issues with using MPI_THREAD_SINGLE while serializing the >> >> > MPI >> >> > calls among the 4 pthreads of each MPI task? That is, is there any >> >> > advantage >> >> > to using MPI_THREAD_FUNELLED or MPI_THREAD_SERIALIZED versus >> >> > MPI_THREAD_SINGLE for serialized calls among pthreads? >> >> > >> >> > Thank you so much, folks. Any help is much appreciated. >> >> > >> >> > Best, >> >> > Thiago. >> >> > >> >> > >> >> > --------------------------------------------------- >> >> > Thiago Quirino, Ph.D. >> >> > NOAA Hurricane Research Division >> >> > 4350 Rickenbacker Cswy. >> >> > Miami, FL 33139 >> >> > P: 305-361-4503 >> >> > E: Thiago.Quirino(a)noaa.gov >> >> > >> >> > _______________________________________________ >> >> > mvapich-discuss mailing list >> >> > mvapich-discuss(a)cse.ohio-state.edu >> >> > http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss >> >> > >> >> >> >> >> >> >> >> -- >> >> Jonathan Perkins >> > >> > >> > >> > _______________________________________________ >> > mvapich-discuss mailing list >> > mvapich-discuss(a)cse.ohio-state.edu >> > http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss >> > >> >> >> >> -- >> Jonathan Perkins >> http://www.cse.ohio-state.edu/~perkinjo >> > -- Jonathan Perkins http://www.cse.ohio-state.edu/~perkinjo

1 0

Re: [mpich-discuss] F90 mpi module issues
by Junchao Zhang 14 Nov '13

14 Nov '13

Hello, Walter, Thanks for reporting that. I created a bug ticket and you are Cc'ed. We hope it can be fixed in the upcoming MPICH 3.1 release soon. --Junchao Zhang On Thu, Nov 14, 2013 at 3:36 PM, W Spector <w6ws(a)earthlink.net> wrote: > The implementation of the MPI module in file mpi_base.f90.in and friends > has serious problems with the various interface blocks. Specifically, with > the exception of the final 'ierror' arguments, the dummy arguments > generally use names of 'v0', 'v1', 'v2', and so on. Arguments should have > the names defined by the report so that callers can use keyword=value style > calls to better document the code. This is a requirement of the MPI 3.0 > Report (section 17.1.3, third bullet), and MPI 2 as well. (Though > admittedly section 16.2.4 of the MPI 2.2 Report is not as explicit in this > regard as the MPI 3.0 Report is.) > > Can this be addressed anytime soon? > > Walter Spector > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss >

1 0

F90 mpi module issues
by W Spector 14 Nov '13

14 Nov '13

The implementation of the MPI module in file mpi_base.f90.in and friends has serious problems with the various interface blocks. Specifically, with the exception of the final 'ierror' arguments, the dummy arguments generally use names of 'v0', 'v1', 'v2', and so on. Arguments should have the names defined by the report so that callers can use keyword=value style calls to better document the code. This is a requirement of the MPI 3.0 Report (section 17.1.3, third bullet), and MPI 2 as well. (Though admittedly section 16.2.4 of the MPI 2.2 Report is not as explicit in this regard as the MPI 3.0 Report is.) Can this be addressed anytime soon? Walter Spector

1 0

Re: [mpich-discuss] Man pages no longer installed
by Ken Raffenetti 14 Nov '13

14 Nov '13

Ah, you are correct on the vpath builds not working. I'll create a bug report to fix this. ----- Original Message ----- > From: "Markus Geimer" <m.geimer(a)fz-juelich.de> > To: discuss(a)mpich.org > Sent: Thursday, November 14, 2013 12:53:13 PM > Subject: Re: [mpich-discuss] Man pages no longer installed > > On 11/14/2013 06:31 PM, Ken Raffenetti wrote: > > To clarify, there are pre-built manpages included in the release > > tarball for 3.1rc1. > > Correct. > > > They are installed as part of 'make install'. > > That's what I expected. Meanwhile I figured out that this is indeed > the case for "regular" in-place builds. But if you do an out-of-place > (aka. VPATH) build, manpages won't be installed. In this case, there > is only a 'doc' subdirectory inside '<prefix>/share'. However, this > should be fixable by using a '$(top_srcdir)' in the corresponding > Makefile rule. > > Thanks, > Markus > > -- > Dr. Markus Geimer > Juelich Supercomputing Centre > Institute for Advanced Simulation > Forschungszentrum Juelich GmbH > 52425 Juelich, Germany > > Phone: +49-2461-61-1773 > Fax: +49-2461-61-6656 > E-Mail: m.geimer(a)fz-juelich.de > WWW: http://www.fz-juelich.de/ias/jsc > > > ------------------------------------------------------------------------------------------------ > ------------------------------------------------------------------------------------------------ > Forschungszentrum Juelich GmbH > 52425 Juelich > Sitz der Gesellschaft: Juelich > Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 > Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher > Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender), > Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt, > Prof. Dr. Sebastian M. Schmidt > ------------------------------------------------------------------------------------------------ > ------------------------------------------------------------------------------------------------ > > _______________________________________________ > 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] Man pages no longer installed
by Markus Geimer 14 Nov '13

14 Nov '13

On 11/14/2013 06:31 PM, Ken Raffenetti wrote: > To clarify, there are pre-built manpages included in the release tarball for 3.1rc1. Correct. > They are installed as part of 'make install'. That's what I expected. Meanwhile I figured out that this is indeed the case for "regular" in-place builds. But if you do an out-of-place (aka. VPATH) build, manpages won't be installed. In this case, there is only a 'doc' subdirectory inside '<prefix>/share'. However, this should be fixable by using a '$(top_srcdir)' in the corresponding Makefile rule. Thanks, Markus -- Dr. Markus Geimer Juelich Supercomputing Centre Institute for Advanced Simulation Forschungszentrum Juelich GmbH 52425 Juelich, Germany Phone: +49-2461-61-1773 Fax: +49-2461-61-6656 E-Mail: m.geimer(a)fz-juelich.de WWW: http://www.fz-juelich.de/ias/jsc ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------ Forschungszentrum Juelich GmbH 52425 Juelich Sitz der Gesellschaft: Juelich Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender), Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt, Prof. Dr. Sebastian M. Schmidt ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------

1 0

Re: [mpich-discuss] Man pages no longer installed
by Antonio J. Peña 14 Nov '13

14 Nov '13

Also, note the configure --mandir option. Antonio On Thursday, November 14, 2013 11:31:07 AM Ken Raffenetti wrote: > To clarify, there are pre-built manpages included in the release tarball for > 3.1rc1. They are installed as part of 'make install'. Are you sure your > MANPATH includes the location of your install? > > Ken > > ----- Original Message ----- > > > From: "Wesley Bland" <wbland(a)mcs.anl.gov> > > To: discuss(a)mpich.org > > Sent: Thursday, November 14, 2013 11:23:39 AM > > Subject: Re: [mpich-discuss] Man pages no longer installed > > > > > > I don’t know when the decision was made to turn off the manages by > > default, but it makes sense as they require a special tool that > > won’t be commonly installed. As the wiki page says, you have to > > install the “sowing” tool first before you can build the man pages. > > If we make that a requirement for the regular `make install`, that > > would be more a burden for most users. > > > > > > Thanks, > > Wesley > > > > > > > > > > On Nov 14, 2013, at 11:17 AM, Junchao Zhang < jczhang(a)mcs.anl.gov > > > wrote: > > > > > > > > Hi, Markus, > > I installed my manpages according to instructions at > > https://wiki.mpich.org/mpich/index.php/Creating_MPICH_Manpages > > It needs a customized tool. > > But whether it is the reason mandoc is not installed by default, I > > would let more experienced mpich developers to answer you. > > > > > > > > --Junchao Zhang > > > > > > On Thu, Nov 14, 2013 at 4:03 AM, Markus Geimer < > > m.geimer(a)fz-juelich.de > wrote: > > > > > > Hi! > > > > Is it intended that MPICH 3.1rc1 no longer installs man pages > > during 'make install'? This seems to be a regression compared > > to MPICH2. > > > > Thanks, > > Markus > > > > -- > > Dr. Markus Geimer > > Juelich Supercomputing Centre > > Institute for Advanced Simulation > > Forschungszentrum Juelich GmbH > > 52425 Juelich, Germany > > > > Phone: +49-2461-61-1773 > > Fax: +49-2461-61-6656 > > E-mail: m.geimer(a)fz-juelich.de > > WWW: http://www.fz-juelich.de/jsc/ > > > > > > > > -------------------------------------------------------------------------- > > ---------------------- > > ------------------------------------------------------------------------- > > ----------------------- Forschungszentrum Juelich GmbH > > 52425 Juelich > > Sitz der Gesellschaft: Juelich > > Eingetragen im Handelsregister des Amtsgerichts Dueren Nr. HR B 3498 > > Vorsitzender des Aufsichtsrats: MinDir Dr. Karl Eugen Huthmacher > > Geschaeftsfuehrung: Prof. Dr. Achim Bachem (Vorsitzender), > > Karsten Beneke (stellv. Vorsitzender), Prof. Dr.-Ing. Harald Bolt, > > Prof. Dr. Sebastian M. Schmidt > > -------------------------------------------------------------------------- > > ---------------------- > > ------------------------------------------------------------------------- > > ----------------------- > > > > _______________________________________________ > > 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 > > > > _______________________________________________ > > 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

1 0