January 2018 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] Building MPICH with Flang
by Halim Amer 15 Jan '18

15 Jan '18

Hi Tarun, Flang is a rather new project. We have tested it some time ago and it failed to build MPICH. Our current stance is that we consider Flang not part of the list of supported compilers (http://www.mpich.org/supported-compilers/) This list might be updated in the future to include Flang but no guarantee and ETA at the moment. Halim http://www.mcs.anl.gov/~aamer On 12/9/17 1:10 AM, Tarun Prabhu wrote: > I have been trying to build MPICH with Clang + Flang. If I build it with > clang without Fortran support, it works fine, but if I try to build it > with Flang for Fortran, I get an unknown argument error in the final > link step. > > clang-5.0: error: unsupported option '--whole-archive' > clang-5.0: error: unsupported option '--no-whole-archive' > clang-5.0: error: unknown argument: '-soname' > clang-5.0: error: no such file or directory: 'libmpi.so.12' > > This seems to be because --whole-archive is passed to clang without > prepending it with -Wl,. So I hacked the libtool script to force it to > add the "-Wl". > > Now I get another error saying: > > usr/bin/ld: src/binding/fortran/mpif_h/.libs/setbotf.o: relocation > R_X86_64_32 against undefined symbol `mpipriv1_' can not be used when > making a shared object; recompile with -fPIC > > Could this be because the build system does not recognize/support flang? > If so, is there a way to work around it? If not, I'd appreciate some > help in trying to get it built. > > Thanks, > > Tarun > > > > _______________________________________________ > 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

Re: [mpich-discuss] MPI_Win_Create
by Jeff Hammond 15 Jan '18

15 Jan '18

Win_create acts like a barrier, so just serialize them like you would any collective. Or even better, don’t do this since there is no benefit. Just create one window. Synchronize where required. In any case, posting code will make it easier for people to help you. Jeff On Mon, Jan 15, 2018 at 6:57 AM Luis Cebamanos <luiceur(a)gmail.com> wrote: > Hi MPI users, > > I am trying to understand the way MPI_Win_Create works. The way > MPI_Win_Create is used in my code creates a window between two MPI > ranks, so it creates a communicator with the two ranks and then proceed > to the create the window. This is also happening in the other direction, > the other rank involved in the communicator creates another window, for > example: > > 0 -> creates a window with 1 > > 1 -> creates a window with 0 > > What I normally do is to play the odd and even game to create the > windows in one direction or the other, but this hangs very often. What > is the right way to do it? Could I serialized the window creations so > there are not deadlocks? > > Regards, > > > _______________________________________________ > discuss mailing list discuss(a)mpich.org > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss > -- Jeff Hammond jeff.science(a)gmail.com http://jeffhammond.github.io/ _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

MPI_Win_Create
by Luis Cebamanos 15 Jan '18

15 Jan '18

Hi MPI users, I am trying to understand the way MPI_Win_Create works. The way MPI_Win_Create is used in my code creates a window between two MPI ranks, so it creates a communicator with the two ranks and then proceed to the create the window. This is also happening in the other direction, the other rank involved in the communicator creates another window, for example: 0 -> creates a window with 1 1 -> creates a window with 0 What I normally do is to play the odd and even game to create the windows in one direction or the other, but this hangs very often. What is the right way to do it? Could I serialized the window creations so there are not deadlocks? 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] mpich-3.2.1 libgfortran.so.3
by Shuwei Zhao 14 Jan '18

14 Jan '18

Hi Bill, So far as I know there are two things you need to check(maybe limited but hope to be helpful): 1. using *chrpath -l* ./*mpiexec* to check the running path of your binary. If you have set it, please delete it. 2. check if the path of libgfortran.so is included in the environmental variable LD_LIBRARY_PATH. The path of the dynamic library link has to be included in the LD_LIBRARY_PATH so that the execution binary can find it. Thanks, Shuwei On Sun, 14 Jan 2018 at 15:24 Robert Sullivan <rjsullivan25(a)gmail.com> wrote: > Hi Bill - I’m no expert, but might be worth double checking that GCC and > gfortran are installed on all the nodes you’re including in the machines > file. > > Rob > > > On Sun, 14 Jan 2018 at 20:27, Bill Petrick <wpetrick(a)industrialimaging.com> > wrote: > >> Hi, >> >> I've installed mpich-v3.2.1 into my cluster but I am having problems. My >> MPICH installation is configured with the gfortran compiler. Although I can >> compile my parallel program with no errors, I can not run it. >> Below is the error message I get. Would you please help me with this >> problem. I also added the following lines to my >> /etc/environment and .bashrc file export PATH=/clusterfiles/mpich2/bin:$PATH >> >> and LD_LIBRARY_PATH=/clusterfiles/mpich2/lib:$LD_LIBRARY_PATH >> >> >> $ mpiexec -f machines -launcher rsh -n 2 ./a.out >> ./a.out: error while loading shared libraries: libgfortran.so.3: cannot >> open shared object file: No such file or directory >> >> libgfortran.so.3 does reside on my computer just not in my nfs'ed >> "/clusterfiles/mpich2/lib" >> >> Please find attached my c.txt, m.txt and mi.txt files >> >> I appreciate any clues you might have. Thanks and cheers, Bill >> >> _______________________________________________ >> 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

Re: [mpich-discuss] mpich-3.2.1 libgfortran.so.3
by Robert Sullivan 14 Jan '18

14 Jan '18

Hi Bill - I’m no expert, but might be worth double checking that GCC and gfortran are installed on all the nodes you’re including in the machines file. Rob On Sun, 14 Jan 2018 at 20:27, Bill Petrick <wpetrick(a)industrialimaging.com> wrote: > Hi, > > I've installed mpich-v3.2.1 into my cluster but I am having problems. My > MPICH installation is configured with the gfortran compiler. Although I can > compile my parallel program with no errors, I can not run it. > Below is the error message I get. Would you please help me with this > problem. I also added the following lines to my > /etc/environment and .bashrc file export > PATH=/clusterfiles/mpich2/bin:$PATH > and LD_LIBRARY_PATH=/clusterfiles/mpich2/lib:$LD_LIBRARY_PATH > > > $ mpiexec -f machines -launcher rsh -n 2 ./a.out > ./a.out: error while loading shared libraries: libgfortran.so.3: cannot > open shared object file: No such file or directory > > libgfortran.so.3 does reside on my computer just not in my nfs'ed > "/clusterfiles/mpich2/lib" > > Please find attached my c.txt, m.txt and mi.txt files > > I appreciate any clues you might have. Thanks and cheers, Bill > > _______________________________________________ > 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

mpich-3.2.1 libgfortran.so.3
by Bill Petrick 14 Jan '18

14 Jan '18

Hi, I've installed mpich-v3.2.1 into my cluster but I am having problems. My MPICH installation is configured with the gfortran compiler. Although I can compile my parallel program with no errors, I can not run it. Below is the error message I get. Would you please help me with this problem. I also added the following lines to my /etc/environment and .bashrc file export PATH=/clusterfiles/mpich2/bin:$PATH and LD_LIBRARY_PATH=/clusterfiles/mpich2/lib:$LD_LIBRARY_PATH $ mpiexec -f machines -launcher rsh -n 2 ./a.out ./a.out: error while loading shared libraries: libgfortran.so.3: cannot open shared object file: No such file or directory libgfortran.so.3 does reside on my computer just not in my nfs'ed "/clusterfiles/mpich2/lib" Please find attached my c.txt, m.txt and mi.txt files I appreciate any clues you might have. Thanks and cheers, Bill _______________________________________________ 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] running make testing
by Bland, Wesley 11 Jan '18

11 Jan '18

You can run `make testing` in any subdirectory. > On Jan 11, 2018, at 12:46 PM, Wei-keng Liao <wkliao(a)eecs.northwestern.edu> wrote: > > As running "make testing" can take a while, I am wondering if > there is a way to test one MPICH "component" at a time? > For example, just to test comm commponent only or io component only. > > Wei-keng > > _______________________________________________ > 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

running make testing
by Wei-keng Liao 11 Jan '18

11 Jan '18

As running "make testing" can take a while, I am wondering if there is a way to test one MPICH "component" at a time? For example, just to test comm commponent only or io component only. Wei-keng _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0

Output files not truncated?
by Joseph Schuchart 10 Jan '18

10 Jan '18

All, While debugging my application using good ol' printf and redirecting stderr and stdout to files using --outfile-pattern/--errfile-pattern, I realized that MPICH does not truncate output files if they already exist. This caused quite a bit of confusion on my side as the interesting information was located at the end of the application's output and I repeatedly found stale data before realizing that the files are not truncated. Is this behavior intended? How to reproduce: $ cat /proc/cpuinfo > test.log $ mpirun --outfile-pattern test.log -n 1 hostname Expected (from me at least): the file (only) contains the output of `hostname`. Result: the file contains the output of `hostname` followed by the remaining content of /proc/cpuinfo. Cheers, Joseph -- Dipl.-Inf. Joseph Schuchart High Performance Computing Center Stuttgart (HLRS) Nobelstr. 19 D-70569 Stuttgart Tel.: +49(0)711-68565890 Fax: +49(0)711-6856832 E-Mail: schuchart(a)hlrs.de _______________________________________________ discuss mailing list discuss(a)mpich.org To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

1 0