[petsc-users] undefined symbol: mpi_init_
Satish Balay
balay at mcs.anl.gov
Sun Jun 26 16:25:14 CDT 2016
On Sun, 26 Jun 2016, Antonio Trande wrote:
> On 06/26/2016 02:24 AM, Satish Balay wrote:
> > On Sat, 25 Jun 2016, Antonio Trande wrote:
> >
> >> On 06/25/2016 09:41 PM, Antonio Trande wrote:
> >>> On 06/25/2016 03:59 PM, Matthew Knepley wrote:
> >>>> On Sat, Jun 25, 2016 at 8:21 AM, Antonio Trande <anto.trande at gmail.com
> >>>> <mailto:anto.trande at gmail.com>> wrote:
> >>>>
> >>>> Hi all,
> >>>>
> >>>> is it possible to compile serial PETSc library ?
> >>>> I'm trying to compile PETSc without MPI by configuring with options:
> >>>>
> >>>> --with-mpi=0 --with-mpiuni-fortran-binding=0
> >>>>
> >>>> but compiler uses -I/.../petsc-3.7.2/include/petsc/mpiuni flag anyway
> >>>> and libpetsc.so contains an "undefined symbol":
> >>>>
> >>>> undefined symbol: mpi_init_ (/usr/lib64/libpetsc.so.3.7.2)
> >>>>
> >>>> How I can fix this issue ?
> >>>> Full log:
> >>>> https://copr-be.cloud.fedoraproject.org/results/sagitter/petsc/fedora-24-x86_64/00362302-petsc/build.log.gz
> >>>>
> >>>>
> >>>> 1) Why are you giving --with-mpiuni-fortran-binding=0 if you still have
> >>>> Fortran turned on? If you want to turn off Fortran, use --with-fc=0.
> >>>
> >>> I did't know what it does. If I active --with-mpiuni-fortran-binding,
> >>> 'undefined symbol' disappears.
> >>>
> >>>>
> >>>> 2) You did not send configure.log which is what I need, not a screen
> >>>> capture.
> >>>>
> >>>> 3) I cannot see the error you sent anywhere. It would be more useful to
> >>>> send the output from 'make test'.
> >>>>
> >>>> Matt
> >>>>
> >>>>
> >>>
> >>> I send you all config.log; petsc-serial, petsc-serial (64bit integer),
> >>> petsc-openmpi and petsc-mpich are compiled from an own fresh source
> >>> directory, respectively from petsc-3.7.2, build64, buildopenmpi_dir,
> >>> buildmpich_dir.
> >>>
> >>> All libraries look compiled fine and related test passed.
> >>> I will send you latest builds log from Fedora Copr build-system.
> >>>
> >>
> >> No, tests not passed with
> >>
> >> [0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation,
> >> probably memory access out of range
> >>
> >> This is a build log from Fedora 24 64 bit:
> >> https://copr-be.cloud.fedoraproject.org/results/sagitter/petsc/fedora-24-x86_64/00362535-petsc/build.log.gz
> >
> >>>>>>>>>
> > + echo 'Patch #1 (petsc-lib64.patch):'
> > + /usr/bin/patch -p0 --fuzz=0
> > patching file config/BuildSystem/config/setCompilers.py
> > patching file gmakefile
> >
> > Patch #0 (petsc-3.7-no-rpath.patch):
> > + /usr/bin/patch -p0 --fuzz=0
> > patching file config/BuildSystem/config/setCompilers.py
> > <<<<<<<<<<<
> >
> > Well looks like there were some modifications done to get the 'multi-variant' build working.
>
> Yes.
>
> >
> >>>>>>>>>>>
> > + make test -C petsc-3.7.2
> > make: Entering directory '/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2'
> > Running test examples to verify correct installation
> > Using PETSC_DIR=/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2 and PETSC_ARCH=x86_64
> > *******************Error detected during compile or link!*******************
> > See http://www.mcs.anl.gov/petsc/documentation/faq.html
> > /builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/src/snes/examples/tutorials ex19
> > *********************************************************************************
> > gcc -o ex19.o -c -O2 -fPIC -O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic -fopenmp -I/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/include -I/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/x86_64/include -I/usr/include -I/usr/include/suitesparse -I/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/include/petsc/mpiuni `pwd`/ex19.c
> > gcc -Wl,-z,relro -O2 -fPIC -O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic -fopenmp -o ex19 ex19.o -L/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/x86_64/lib -L/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/x86_64/lib -lpetsc -lmetis -L/usr/lib64 -lumfpack -lklu -lcholmod -lbtf -lccolamd -lcolamd -lcamd -lamd -lsuitesparseconfig -llapack -lblas -lOpenCL -lhwloc -lpapi -lgmp -lpthread -ljpeg -lpng -lm -L/usr/lib/gcc/x86_64-redhat-linux/6.1.1 -lgfortran -lm -lgfortran -lm -lquadmath -lm -lstdc++ -L/usr/lib/gcc/x86_64-redhat-linux/6.1.1 -ldl -lgcc_s -ldl
> > /usr/bin/rm -f ex19.o
> > Possible error running C/C++ src/snes/examples/tutorials/ex19 with 1 MPI process
> > See http://www.mcs.anl.gov/petsc/documentation/faq.html
> > [0]PETSC ERROR: #1 PetscLogInitialize() line 219 in /builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/src/sys/logging/plog.c
> > [0]PETSC ERROR: #2 PetscInitialize() line 891 in /builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2/src/sys/objects/pinit.c
> > /usr/bin/sh: line 13: 16933 Aborted (core dumped)
> > <<<<<<<<<<
> >
> > So there is no -rpath here. Which -lpetsc variant is getting used at runtime?
> >
>
> Yes, no rpaths for the serial libraries, so 'libpetsc.so' and
> 'libpetsc64.so' (renamed for mine needs).
>
> Test reported is for PETSc (32 bit integer) serial library: libpetsc.so
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
+ mkdir -p /builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/lib64 /builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/share/petsc /builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/include/petsc
+ mkdir -p /builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/lib64/gfortran/modules/petsc-3.7.2 /builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/share/Modules/modulefiles
+ install -pm 755 x86_64/lib/libpetsc.so x86_64/lib/libpetsc.so.3.7 x86_64/lib/libpetsc.so.3.7.2 /builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/lib64
+ umask 022
+ cd /builddir/build/BUILD
+ cd petsc-3.7.2
+ export LD_LIBRARY_PATH=/usr/lib64:/builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/lib64
+ LD_LIBRARY_PATH=/usr/lib64:/builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/lib64
+ export PETSC_DIR=/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2
+ PETSC_DIR=/builddir/build/BUILD/petsc-3.7.2/petsc-3.7.2
+ export PETSC_ARCH=x86_64
+ PETSC_ARCH=x86_64
+ make test -C petsc-3.7.2
<<<<<<<<<<<<<<<<<<<<<<<<<<<<<
Ok - I see libpetsc.so is copied over to
/builddir/build/BUILDROOT/petsc-3.7.2-2.fc24.x86_64/usr/lib64. And
this path is set in LD_LIBRARY_PATH - so the correct -lpetsc should be
found at runtime for 'make test'
I attempted to do a build similar to what done here.
>>>>>>>>>>>>>>>>>>>
$ cat x86_64.py
#!/usr/bin/python
if __name__ == '__main__':
import sys
import os
sys.path.insert(0, os.path.abspath('config'))
import configure
configure_options = [
'--prefix=/usr',
'CFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic',
'CXXFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic',
'FFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic',
'LDFLAGS=-Wl,-z,relro -specs=/usr/lib/rpm/redhat/redhat-hardened-ld',
'--with-default-arch=0',
'--with-make=1',
'--with-single-library=1',
'--with-precision=double',
'--with-petsc-arch=x86_64',
'--with-clanguage=C',
'--with-shared-libraries=1',
'--with-64-bit-indices=0',
'--with-fortran-interfaces=1',
'--with-windows-graphics=0',
'--CC=gcc',
'--CXX=g++',
'--FC=gfortran',
'--with-gnu-compilers=1',
'--with-vendor-compilers=0',
'--with-shared-ld=ld',
'--with-pic=1',
'--sharedLibraryFlags=',
'--dynamicLibraryFlags=',
'--with-cxx-dialect=',
'--with-debugging=0',
'COPTFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic',
'CXXOPTFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic',
'FOPTFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic -I/usr/lib64/gfortran/modules',
'--with-blas-lib=libblas.so',
'--with-lapack-lib=liblapack.so',
'--with-blas-lapack-suffix=',
'--known-64-bit-blas-indices=0',
'--with-mpi=0',
'--with-mpiuni-fortran-binding=1',
'--with-pthread=1',
'--with-valgrind=1',
'--with-valgrind-dir=/usr',
]
configure.petsc_configure(configure_options)
<<<<<<<<<<<
$
But I can't reprodue this error - with the above build.. The suggestions I have for debugging are:
Somehow get the following through the build - and verify if the includes, libraries are picked up correctly
cd src/snes/examples/tutorials
make PETSC_DIR=/home/balay/petsc PETSC_ARCH=x86_64 ex19
ldd ex19 |grep libpetsc
<verify the correct libpetsc.so is used>
make PETSC_DIR=/home/balay/petsc PETSC_ARCH=x86_64 ex19.o CFLAGS=-E
grep petscconf.h ex19.o
<verify the correct petscconf.h is used hre>
Or you could somehow try to do a minimal build - and see if the problem is reproduceable.
[say just build 1 variant of PETSc, fewer externalpackages].
BTW: you have:
++ sed -e s/-D_FORTIFY_SOURCE/-D-FORTIFY-SOURCE/g
CFLAGS=-O2 -g -pipe -Wall -Werror=format-security -Wp,-D-FORTIFY-SOURCE=2 -fexceptions -fstack-protector-strong --param=ssp-buffer-size=4 -grecord-gcc-switches -specs=/usr/lib/rpm/redhat/redhat-hardened-cc1 -m64 -mtune=generic
However -Wp,-D-FORTIFY-SOURCE=2 gives me compile errors.
$ gcc -Wp,-D-FORTIFY-SOURCE=2 conftest.c
<command-line>:0:1: error: macro names must be identifiers
$
Satish
More information about the petsc-users
mailing list