Page 1 of 1
Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_attr_ge
Posted: Mon Apr 02, 2012 4:16 pm
by scanya
Hello,
I have a problem with run abinit from pbs server.
I have compiled it this way:
export FC=ifort
export CC=icc
./configure --prefix=/opt/exp_soft/plgrid/abinit-mpi-intel
Every time I submit pbs job I get such error:
MPIFAKE: mpi_attr_get
MPIFAKE: mpi_attr_get
MPIFAKE: mpi_attr_get
MPIFAKE: mpi_attr_get
here is my pbsjob:
Code: Select all
#PBS -N abinit
#PBS -q default
#PBS -j oe
#PBS -l nodes=1:ppn=8:ib
#PBS -l walltime=10:00:00
#PBS -V
export MPI_VERSION="openmpi_intel"
source /etc/mpi.sh
cd $PBS_O_WORKDIR
$MPI_HOME/bin/mpiexec /opt/exp_soft/plgrid/abinit-mpi-intel/bin/abinit -stdin tparal_1.files
Any idea what could be wrong?
---
Cheers,
Bartek
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Mon Apr 02, 2012 6:17 pm
by pouillon
You didn't enable MPI when building Abinit. That's why you cannot run it in parallel.
Try the following, without setting CC nor FC:
Code: Select all
../configure --enable-mpi --enable-mpi-io --with-mpi-prefix=/path/to/your/mpi/installation
The build system of Abinit will expect to find mpicc and mpif90 in /path/to/your/mpi/installation/bin/.
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Tue Apr 03, 2012 2:51 pm
by scanya
Thank you for your answer.
If I run that:
./configure --enable-mpi --enable-mpi-io --with-mpi-prefix=/usr/mpi/intel/openmpi --prefix=/opt/exp_soft/plgrid/abinit
it finishes ok
but make
cause:
Code: Select all
make[6]: *** [atompaw] Error 1
make[6]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks/sources/atompaw-3.0.1.3/src'
make[5]: *** [all-recursive] Error 1
make[5]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks/sources/atompaw-3.0.1.3'
make[4]: *** [all] Error 2
make[4]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks/sources/atompaw-3.0.1.3'
make[3]: *** [stamps/atompaw-build-stamp] Error 2
make[3]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks'
make[2]: [all-local] Error 2 (ignored)
Checking build of atompaw fallback
test -e stamps/atompaw-install-stamp
make[2]: *** [all-local] Error 1
make[2]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2'
make: *** [all] Error 2
:/ What else could be wrong?
---
Cheers,
Bartek
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Tue Apr 03, 2012 5:35 pm
by pouillon
The useful error message is just a few lines above what you cut and pasted. Without it, we cannot tell you much.
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Wed Apr 04, 2012 10:18 am
by scanya
Hello,
My fault .. here is the full error:
Code: Select all
ifort: command line remark #10010: option '-pthread' is deprecated and will be removed in a future release. See '-help deprecated'
ifort: command line warning #10156: ignoring option '-r'; no argument required
ifort: command line warning #10156: ignoring option '-r'; no argument required
ipo: warning #11010: file format not recognized for /usr/mpi/intel/openmpi-1.4.3/lib64
/usr/mpi/intel/openmpi-1.4.3/lib64: file not recognized: Is a directory
make[6]: *** [atompaw] Error 1
make[6]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks/sources/atompaw-3.0.1.3/src'
make[5]: *** [all-recursive] Error 1
make[5]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks/sources/atompaw-3.0.1.3'
make[4]: *** [all] Error 2
make[4]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks/sources/atompaw-3.0.1.3'
make[3]: *** [stamps/atompaw-build-stamp] Error 2
make[3]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks'
make[2]: [all-local] Error 2 (ignored)
Checking build of atompaw fallback
test -e stamps/atompaw-install-stamp
make[2]: *** [all-local] Error 1
make[2]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2/fallbacks'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/admins/xxxx/name/abinit/abinit-6.12.2'
make: *** [all] Error 2
TIA,
Bartek
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Tue Apr 17, 2012 5:31 pm
by pouillon
You may safely disable AtomPAW, BigDFT and Wannier90 support, unless you really need them. The following should solve your problem:
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Wed May 09, 2012 11:43 am
by scanya
OK. I get through it. I have to use libxc, and now after such configure:
Code: Select all
./configure CC=icc FC=mpif90 --enable-mpi=yes --with-linalg-flavor=mkl --enable-64bit-flags=yes --with-linalg-libs=-L/opt/intel/mkl/lib/intel64 --enable-libxc --with-atompaw-libs=-L/usr/lib64 --with-atompaw-bins=/usr/bin --prefix=/opt/exp_soft/xxx/abinit
I'm getting:
configure: WARNING: falling back to internal libxc version
checking whether to select a fallback for atompaw... no
configure: WARNING: AtomPAW requires missing linear algebra support
configure: WARNING: AtomPAW requires missing LibXC support
+--------------------------------------------------------------+
| Connector detection failure |
+--------------------------------------------------------------+
| The build system of Abinit has not been able to detect |
| one or more requested external packages, probably because |
| you did not provide sufficient information. This means |
| that the build cannot proceed, and the configure script |
| will thus abort now with an error. |
| |
| To address this issue, you may: |
| |
| * make sure that the prerequisites of the external |
| packages are available and enabled, e.g. |
| --with-dft-flavor="libxc+bigdft" or |
| --with-trio-flavor="netcdf+etsf_io" |
| |
| * use the --with-*-bins, --with-*-incs, and |
| --with-*-libs options of configure to provide |
| additional information; please see the section related |
| to connectors in |
| ~abinit/doc/config/build-config.ac for available |
| options and details; |
| |
| * disable the corresponding extensions, by using the |
| associated --with-*-flavor="none" options of |
| configure; please see the section associated to |
| connectors in ~abinit/doc/config/build-config.ac for |
| details. |
| |
+--------------------------------------------------------------+
configure: error: external atompaw support does not work
I've build atompaw externally:
Code: Select all
./configure --prefix=/usr --enable-mpi CC=gcc --with-linalg-flavor=mkl --with-linalg-libs="-lmkl_blacs_lp64 -lmkl_intel_thread -lmkl_core -liomp5 -lpthread -lmkl_intel_lp64 -lmkl_sequential" --enable-libxc --with-libxc-incs="-I/usr/lib64/gfortran/modules -I/usr/include"
Thank you for help,
Bartek
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Wed May 09, 2012 5:23 pm
by pouillon
AtomPAW is not necessary to Abinit, and does not provide any extra feature. You may safely disable it (see the --with-dft-flavor option).
It would anyway be a good idea to provide us with the config.log of Abinit, so that we can check that this is not coming from a bug in the build system. Thank you in advance.
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Thu May 10, 2012 4:31 pm
by scanya
Hello,
Thank you for the answer. I have to compile it with a linear algebra.
Please check attachement there is a config.log
---
Cheers,
Bartek
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Thu May 10, 2012 5:36 pm
by pouillon
Looks like --with-libxc-incs does not point to the LibXC modules.
The --enable-libxc option is not recognized by the Abinit build system. You should use --with-dft-flavor="libxc" instead.
Please look into ~abinit/doc/config/build-config.ac for more details.
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Mon May 14, 2012 10:18 am
by scanya
Hello,
Thank you for help.
Now everything seems to be fine ... almost fine.
It compiles perfectly but after
make, when I run
make check it finishes with such error:
Code: Select all
[libxc][t999] File t*_SUS* erased
[libxc][t999] File t*_KSS* erased
[libxc][t999] File t*_xsf* erased
make[4]: Leaving directory `/tmp/scanya/abinit-6.12.3/tests'
make[3]: Leaving directory `/tmp/scanya/abinit-6.12.3/tests'
etsf_io/tmp-test_Linux_20120511/report:Case_06 failed
etsf_io/tmp-test_Linux_20120511/report:Case_07 failed
etsf_io/tmp-test_Linux_20120511/report:Case_08 failed
tutorespfn/tmp-test_Linux_20120511/report:Case_elast_5 failed (too many erroneous lines : 2 , accepted 1 )
tutorespfn/tmp-test_Linux_20120511/report:Case_nlo_5 failed (too large absolute error : 0.008489203 , accepted 0.000e+00 )
tutorial/tmp-test_Linux_20120511/report:Case_gw1_4 failed (too many erroneous lines : 23 , accepted 20 )
tutorial/tmp-test_Linux_20120511/report:Case_gw1_5 failed (too many erroneous lines : 14 , accepted 10 )
tutorial/tmp-test_Linux_20120511/report:Case_gw1_8 failed (too many erroneous lines : 31 , accepted 27 )
v2/tmp-test_Linux_20120511/report:Case_13 failed
v2/tmp-test_Linux_20120511/report:Case_18 failed
v2/tmp-test_Linux_20120511/report:Case_21 failed (too large absolute error : 0.00359935108538 , accepted 2.510e-10 )
v4/tmp-test_Linux_20120511/report:Case_54 failed (too large absolute error : 0.00999999999999091 , accepted 0.000e+00 )
v4/tmp-test_Linux_20120511/report:Case_68 failed (too large absolute error : 0.00457609999999997 , accepted 5.637e-05 )
v4/tmp-test_Linux_20120511/report:Case_71 failed
v5/tmp-test_Linux_20120511/report:Case_45 failed
v6/tmp-test_Linux_20120511/report:Case_37 failed (too large absolute error : 0.6677935615 , accepted 2.0e-5 )
v6/tmp-test_Linux_20120511/report:Case_40 failed (too large absolute error : 2.09999999993826e-05 , accepted 2.0e-7 )
v6/tmp-test_Linux_20120511/report:Case_89 failed (too large relative error : 1 , accepted 6.0e-6 )
make[2]: *** [check-local] Error 1
make[2]: Leaving directory `/tmp/scanya/abinit-6.12.3/tests'
make[1]: *** [check-am] Error 2
make[1]: Leaving directory `/tmp/scanya/abinit-6.12.3/tests'
make: *** [check-recursive] Error 1
Any idea what else could be wrong?
--
Thank you,
Bartek
Re: Scientific Linux CERN SLC release 5.7 - MPIFAKE: mpi_att
Posted: Mon May 14, 2012 2:57 pm
by pouillon
The final accuracy of your results highly depends on the optimizations you enable at build-time and cannot always be predicted. If you want to evaluate the influence of optimization flags for your system, just rebuild Abinit with less optimizations (e.g. FCFLAGS_OPTIM="-g -O2") and re-run the test suite.