July 2021 - discuss - mailman3.cels.anl.gov

Re: [mpich-discuss] Build failed with Undefined symbols
by Raffenetti, Kenneth J. 20 Jul '21

20 Jul '21

Issue was the configure version check only applied to MPICH and not Hydra. Pushed a fix in https://github.com/pmodels/mpich/pull/5449. Ken On 7/20/21, 10:28 AM, "Raffenetti, Kenneth J. via discuss" <discuss(a)mpich.org> wrote: We did add a minimum version check during configure, but it may have fallen out of date. We can take a look. Ken On 7/20/21, 8:18 AM, "Junchao Zhang via discuss" <discuss(a)mpich.org> wrote: Adding --with-hwloc=embedded fixed the problem. Yes, I have preinstalled hwloc (it is common since I also found hwloc on another Linux cluster that had the same kind of hw errors). I guess it was because mpich is not compatible with the system's preinstalled hwloc.Perhaps a better approach is for mpich to check the hwloc at configure time (e.g., many petsc third party libraries are checked in this way), then report any compatibility issue and give clear messages to users asking them to either upgrade their hwloc or use the embedded one. --Junchao Zhang On Mon, Jul 19, 2021 at 6:16 PM Tony Curtis <anthony.curtis(a)stonybrook.edu> wrote: > On Jul 19, 2021, at 3:59 PM, Junchao Zhang via discuss <discuss(a)mpich.org> wrote: > > Hello, > > I configured mpich/main (or 4.0a2) with > ./configure CC=gcc CXX=g++ --enable-fortran=no > on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build failed with > > ... > CCLD hydra_pmi_proxy > Undefined symbols for architecture x86_64: > "_hwloc_topology_set_io_types_filter", referenced from: > _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) > Do you have hwloc installed (e.g. via brew)? Or if so, is it new enough? Tony _______________________________________________ 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] Build failed with Undefined symbols
by Raffenetti, Kenneth J. 20 Jul '21

20 Jul '21

We did add a minimum version check during configure, but it may have fallen out of date. We can take a look. Ken On 7/20/21, 8:18 AM, "Junchao Zhang via discuss" <discuss(a)mpich.org> wrote: Adding --with-hwloc=embedded fixed the problem. Yes, I have preinstalled hwloc (it is common since I also found hwloc on another Linux cluster that had the same kind of hw errors). I guess it was because mpich is not compatible with the system's preinstalled hwloc.Perhaps a better approach is for mpich to check the hwloc at configure time (e.g., many petsc third party libraries are checked in this way), then report any compatibility issue and give clear messages to users asking them to either upgrade their hwloc or use the embedded one. --Junchao Zhang On Mon, Jul 19, 2021 at 6:16 PM Tony Curtis <anthony.curtis(a)stonybrook.edu> wrote: > On Jul 19, 2021, at 3:59 PM, Junchao Zhang via discuss <discuss(a)mpich.org> wrote: > > Hello, > > I configured mpich/main (or 4.0a2) with > ./configure CC=gcc CXX=g++ --enable-fortran=no > on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build failed with > > ... > CCLD hydra_pmi_proxy > Undefined symbols for architecture x86_64: > "_hwloc_topology_set_io_types_filter", referenced from: > _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) > Do you have hwloc installed (e.g. via brew)? Or if so, is it new enough? Tony

1 0

Re: [mpich-discuss] Build failed with Undefined symbols
by Junchao Zhang 20 Jul '21

20 Jul '21

Adding --with-hwloc=embedded fixed the problem. Yes, I have preinstalled hwloc (it is common since I also found hwloc on another Linux cluster that had the same kind of hw errors). I guess it was because mpich is not compatible with the system's preinstalled hwloc. Perhaps a better approach is for mpich to check the hwloc at configure time (e.g., many petsc third party libraries are checked in this way), then report any compatibility issue and give clear messages to users asking them to either upgrade their hwloc or use the embedded one. --Junchao Zhang On Mon, Jul 19, 2021 at 6:16 PM Tony Curtis <anthony.curtis(a)stonybrook.edu> wrote: > > > > On Jul 19, 2021, at 3:59 PM, Junchao Zhang via discuss < > discuss(a)mpich.org> wrote: > > > > Hello, > > > > I configured mpich/main (or 4.0a2) with > > ./configure CC=gcc CXX=g++ --enable-fortran=no > > on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build > failed with > > > > ... > > CCLD hydra_pmi_proxy > > Undefined symbols for architecture x86_64: > > "_hwloc_topology_set_io_types_filter", referenced from: > > _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) > > > > Do you have hwloc installed (e.g. via brew)? Or if so, is it new enough? > > Tony > >

1 0

Re: [mpich-discuss] Build failed with Undefined symbols
by Tony Curtis 19 Jul '21

19 Jul '21

> On Jul 19, 2021, at 3:59 PM, Junchao Zhang via discuss <discuss(a)mpich.org> wrote: > > Hello, > > I configured mpich/main (or 4.0a2) with > ./configure CC=gcc CXX=g++ --enable-fortran=no > on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build failed with > > ... > CCLD hydra_pmi_proxy > Undefined symbols for architecture x86_64: > "_hwloc_topology_set_io_types_filter", referenced from: > _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) > Do you have hwloc installed (e.g. via brew)? Or if so, is it new enough? Tony

1 0

Re: [mpich-discuss] Build failed with Undefined symbols
by Zhou, Hui 19 Jul '21

19 Jul '21

Hi Junchao, Not sure what happened. Do you have the output log for configure and make? Hui ________________________________ From: Junchao Zhang via discuss <discuss(a)mpich.org> Sent: Monday, July 19, 2021 2:59 PM To: discuss(a)mpich.org <discuss(a)mpich.org> Cc: Junchao Zhang <junchao.zhang(a)gmail.com> Subject: [mpich-discuss] Build failed with Undefined symbols Hello, I configured mpich/main (or 4.0a2) with ./configure CC=gcc CXX=g++ --enable-fortran=no on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build failed with ... CCLD hydra_pmi_proxy Undefined symbols for architecture x86_64: "_hwloc_topology_set_io_types_filter", referenced from: _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) --Junchao Zhang

1 0

Build failed with Undefined symbols
by Junchao Zhang 19 Jul '21

19 Jul '21

Hello, I configured mpich/main (or 4.0a2) with ./configure CC=gcc CXX=g++ --enable-fortran=no on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build failed with ... CCLD hydra_pmi_proxy Undefined symbols for architecture x86_64: "_hwloc_topology_set_io_types_filter", referenced from: _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) --Junchao Zhang

1 0

Build failed with Undefined symbols
by Junchao Zhang 19 Jul '21

19 Jul '21

Hello, I configured mpich/main (or 4.0a2) with ./configure CC=gcc CXX=g++ --enable-fortran=no on Mac with Apple clang version 12.0.5 (clang-1205.0.22.11). The build failed with ... CCLD hydra_pmi_proxy Undefined symbols for architecture x86_64: "_hwloc_topology_set_io_types_filter", referenced from: _HYDT_topo_hwloc_init in libhydra.a(topo_hwloc.o) --Junchao Zhang

1 0

Re: [mpich-discuss] Trying to work with the MPICH test suite, need a little help processing the results
by Zhou, Hui 14 Jul '21

14 Jul '21

The output XML is in JUNIT format. I trust you can easily look up the schema by search for JUNIT. The current runtests script doesn’t handle tests with paths. Rather, it is designed to work recursively. For example, the top testlist should only contain direct subdirectory, e.g. rma. Then runtests will look for separate testlist file in rma/. What you need to do is to edit `rma/testlist` so it only contains `sampletest`. -- Hui Zhou From: Heinz, Michael William via discuss <discuss(a)mpich.org> Date: Wednesday, July 14, 2021 at 7:49 AM To: discuss(a)mpich.org <discuss(a)mpich.org> Cc: Heinz, Michael William <michael.william.heinz(a)cornelisnetworks.com> Subject: [mpich-discuss] Trying to work with the MPICH test suite, need a little help processing the results Hey, So… I’m working with the MPICH test suite to check for bugs in some pre-release OFI changes I’m working on, but I could use a little help processing the resulting XML files. First – is there a schema file for the output XML files? (I know about TestResults.xsl but I’m hoping to generate some Excel pivot tables for my managers to view) Second – I figured out how to exclude individual tests by creating my own testlist files but when I do, the XML output loses information about what type of test is being run. For example, if my test list reads: # do the rma tests rma Then the output will have the form: <MPITEST> <NAME>sampletest</NAME> <ARGS></ARGS> <NP>2</NP> <WORKDIR>./rma</WORKDIR> <STATUS>pass</STATUS> <TIME>0.123456789</TIME> </MPITEST> But if my test list reads: # do rma/sampletest rma/sampletest then the output will be: <MPITEST> <NAME>sampletest</NAME> <ARGS></ARGS> <NP>2</NP> <WORKDIR>./</WORKDIR> <STATUS>pass</STATUS> <TIME>0.123456789</TIME> </MPITEST> This makes it a bit harder to quickly group error reports by category. Any suggestions for retaining the working directory in this case? Thanks in advance. --- Michael Heinz Fabric Software Engineer, Cornelis Networks External recipient

1 0

Trying to work with the MPICH test suite, need a little help processing the results
by Heinz, Michael William 14 Jul '21

14 Jul '21

Hey, So... I'm working with the MPICH test suite to check for bugs in some pre-release OFI changes I'm working on, but I could use a little help processing the resulting XML files. First - is there a schema file for the output XML files? (I know about TestResults.xsl but I'm hoping to generate some Excel pivot tables for my managers to view) Second - I figured out how to exclude individual tests by creating my own testlist files but when I do, the XML output loses information about what type of test is being run. For example, if my test list reads: # do the rma tests rma Then the output will have the form: <MPITEST> <NAME>sampletest</NAME> <ARGS></ARGS> <NP>2</NP> <WORKDIR>./rma</WORKDIR> <STATUS>pass</STATUS> <TIME>0.123456789</TIME> </MPITEST> But if my test list reads: # do rma/sampletest rma/sampletest then the output will be: <MPITEST> <NAME>sampletest</NAME> <ARGS></ARGS> <NP>2</NP> <WORKDIR>./</WORKDIR> <STATUS>pass</STATUS> <TIME>0.123456789</TIME> </MPITEST> This makes it a bit harder to quickly group error reports by category. Any suggestions for retaining the working directory in this case? Thanks in advance. --- Michael Heinz Fabric Software Engineer, Cornelis Networks External recipient

1 0

Re: [mpich-discuss] How to configure mpich without cuda
by David Alexander 13 Jul '21

13 Jul '21

We were using 3.4.1, which I guess you can tell from the directory. I will try 3.4.2. Thanks! Dave > On Jul 13, 2021, at 9:45 AM, Raffenetti, Kenneth J. <raffenet(a)mcs.anl.gov> wrote: > > MPICH 3.4.2 should support --without-cuda, too. 3.4.1 does not. > > Ken > > On 7/13/21, 9:51 AM, "Zhou, Hui via discuss" <discuss(a)mpich.org <mailto:[email protected]>> wrote: > > Hi David, > > The fix probably didn’t go into the 3.4 release. If you could try the 4.0a2 release, `--without-cuda` should work. > > -- > Hui Zhou > > > > > > From: David Alexander <alexanda(a)txcorp.com> > Date: Tuesday, July 13, 2021 at 12:21 AM > To: Zhou, Hui <zhouh(a)anl.gov> > Cc: discuss(a)mpich.org <discuss(a)mpich.org>, John Cary <cary(a)colorado.edu> > Subject: Re: [mpich-discuss] How to configure mpich without cuda > > The “—without-cuda" option causes the compile to fail (where without it the compile succeeds)... > > > > > [qar@qacentos7 shared]$ make 2>&1 | tee build.txt > > make all-recursive > > make[1]: Entering directory `/scr_qacentos7/qar/jenkins/workspace/txliball/n/qacentos7/builds-v/mpich-3.4.1/shared' > > Making all in src/mpl > > make[2]: Entering directory `/scr_qacentos7/qar/jenkins/workspace/txliball/n/qacentos7/builds-v/mpich-3.4.1/shared/src/mpl' > > CCLD libmpl.la <http://libmpl.la/> <http://libmpl.la <http://libmpl.la/>> > > ./libtool: line 7472: cd: no/lib: No such file or directory > > libtool: error: cannot determine absolute directory name of 'no/lib' > > make[2]: *** [libmpl.la <http://libmpl.la/> <http://libmpl.la <http://libmpl.la/>>] Error 1 > > make[2]: Leaving directory `/scr_qacentos7/qar/jenkins/workspace/txliball/n/qacentos7/builds-v/mpich-3.4.1/shared/src/mpl' > > make[1]: *** [all-recursive] Error 1 > > make[1]: Leaving directory `/scr_qacentos7/qar/jenkins/workspace/txliball/n/qacentos7/builds-v/mpich-3.4.1/shared' > > make: *** [all] Error 2 > > > > > > > > > > On Jul 12, 2021, at 8:24 PM, Zhou, Hui <zhouh(a)anl.gov <mailto:[email protected]>> wrote: > > > Hi John, > > > > Could you try configure mpich with `--without-cuda`? > > ________________________________________ > > From: John Cary via discuss <discuss(a)mpich.org <mailto:[email protected]>> > Sent: Monday, July 12, 2021 6:27 PM > To: discuss(a)mpich.org <mailto:[email protected]> <discuss(a)mpich.org <mailto:[email protected]>>; David Alexander <alexanda(a)txcorp.com <mailto:[email protected]>> > Cc: John Cary <cary(a)colorado.edu <mailto:[email protected]>> > Subject: [mpich-discuss] How to configure mpich without cuda > > > > > Right now it picks up cuda, which links libcuda.so into mpiexec. Is there > any way to prevent this? > > Thx....John Cary > _______________________________________________ > discuss mailing list discuss(a)mpich.org <mailto:[email protected]> > To manage subscription options or unsubscribe: > https://lists.mpich.org/mailman/listinfo/discuss <https://lists.mpich.org/mailman/listinfo/discuss>

1 0