parallel error

Jordan · Post by **Jordan** » Thu Apr 02, 2015 9:35 am

You can put all thos variable in your build.ac if you'd like
Remove all the "--" before "with" and replace all "-" with "_" like the following

Code: Select all

FC=mpif90
CC=mpicc
CXX=mpicxx
enable_mpi=yes
with_mpi_prefix="/your/path/to/mpi/installation/dir"
FCFLAGS_EXTRA="-traceback"
with_linalg_flavor="mkl"
with_linalg_incs="-I/opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64" 
with_linalg_libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group"

and so on.
Note there is a typo in your line for --with-linalg_libs, it should have been --with-linalg-libs

Now run ./configure --with-config-file=build.ac

Post by **jbeuken** » Fri Apr 03, 2015 10:29 am

Hi

if use this command line:
./configure \
--with-linalg-flavor="mkl" \
--with-linalg-incs="-I$opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64" \
--with-linalg_libs="-L$opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group" \
--with-fft-flavor=fftw3 \
--with-fft-libs="-L$mhmood/fftw-3.3.4 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group \
"FCFLAGS_EXTRA="-traceback"

4 remarks :
- replace all "$opt" by "/opt"
- for mpi support , you need to add : --enable-mpi --with-mpi-prefix="/usr/local/openmpi" ( if OpenMPI is installed in /usr/local/openmpi ! )
- with your "--with-linalg_libs" definition , you must use ifort !
if you use gfortran , you must use "-lmkl_gf_lp64" instead of "-lmkl_intel_lp64"
- as you have mkl,it's better to use fftw3 from mkl

then ,

Code: Select all

./configure  \
--enable-mpi \
--with-mpi-prefix="/usr/local/openmpi" \
--with-linalg-flavor="mkl"  \
--with-linalg-incs="-I/opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64"  \
--with-linalg_libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group" \
--with-fft-flavor=fftw3-mkl \
--with-fft-libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group"

jmb

hammouri · Post by **hammouri** » Sun Apr 05, 2015 7:50 am

Jordan & jbeuken thank you!
jbeuken, one question:
Do I need to replace "-lmkl_intel_lp64" with "lmkl_gf_lp64" for --with-fft-libs?
I already did and I will test it and let you know.

MH

hammouri · Post by **hammouri** » Sun Apr 05, 2015 8:31 am

Based on my little experience on compilation, a command lines like the below one:
--with-linalg-incs="-I/opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64"
assumes I must have a file called lining-incs and also to have linalg_libs ... etc.
I just went to the folder : /opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64, and all of the file I found are:
libmkl_avx.so* libmkl_intel_sp2dp.so*
libmkl_blacs_ilp64.a libmkl_intel_thread.a
libmkl_blacs_intelmpi_ilp64.a libmkl_intel_thread.so*
libmkl_blacs_intelmpi_ilp64.so* libmkl_lapack95_ilp64.a
libmkl_blacs_intelmpi_lp64.a libmkl_lapack95_lp64.a
libmkl_blacs_intelmpi_lp64.so* libmkl_mc3.so*
libmkl_blacs_lp64.a libmkl_mc.so*
libmkl_blacs_openmpi_ilp64.a libmkl_p4n.so*
libmkl_blacs_openmpi_lp64.a libmkl_pgi_thread.a
libmkl_blacs_sgimpt_ilp64.a libmkl_pgi_thread.so*
libmkl_blacs_sgimpt_lp64.a libmkl_rt.so*
libmkl_blas95_ilp64.a libmkl_scalapack_ilp64.a
libmkl_blas95_lp64.a libmkl_scalapack_ilp64.so*
libmkl_cdft_core.a libmkl_scalapack_lp64.a
libmkl_cdft_core.so* libmkl_scalapack_lp64.so*
libmkl_core.a libmkl_sequential.a
libmkl_core.so* libmkl_sequential.so*
libmkl_def.so* libmkl_solver_ilp64.a
libmkl_gf_ilp64.a libmkl_solver_ilp64_sequential.a
libmkl_gf_ilp64.so* libmkl_solver_lp64.a
libmkl_gf_lp64.a libmkl_solver_lp64_sequential.a
libmkl_gf_lp64.so* libmkl_vml_avx.so*
libmkl_gnu_thread.a libmkl_vml_def.so*
libmkl_gnu_thread.so* libmkl_vml_mc2.so*
libmkl_intel_ilp64.a libmkl_vml_mc3.so*
libmkl_intel_ilp64.so* libmkl_vml_mc.so*
libmkl_intel_lp64.a libmkl_vml_p4n.so*
libmkl_intel_lp64.so* locale/
libmkl_intel_sp2dp.a
and also no file like fft-libs

MH

Post by **jbeuken** » Sun Apr 05, 2015 8:43 pm

if you use gfortran ( than GNU GCC but with mkl ) then use

Code: Select all

--with-linalg-flavor="mkl"  \
--with-linalg-incs="-I/opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64"  \
--with-linalg_libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group" \
--with-fft-flavor=fftw3-mkl \
--with-fft-libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group"

the fftw3 calls are included in mkl library... 8-)

but if you use ifort , then you use :

Code: Select all

--with-linalg-flavor="mkl"  \
--with-linalg-incs="-I/opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64"  \
--with-linalg_libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group" \
--with-fft-flavor=fftw3-mkl \
--with-fft-libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_intel_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group"

jmb

hammouri · Post by **hammouri** » Mon Apr 06, 2015 9:01 pm

As I have gfortran, I did this:

./configure \
> --enable-mpi \
> --with-mpi-prefix="/usr/local/openmpi" \
> --with-linalg-flavor="mkl" \
> --with-linalg-incs="-I/opt/intel/composer_xe_2011_sp1.10.319/mkl/include/intel64/lp64" \
> --with-linalg_libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group" \
> --with-fft-flavor=fftw3-mkl \
> --with-fft-libs="-L/opt/intel/composer_xe_2011_sp1.10.319/mkl/lib/intel64 -Wl,--start-group -lmkl_gf_lp64 -lmkl_sequential -lmkl_core -Wl,--end-group"

and this is the summary:

Summary of important options:

* C compiler : gnu version 4.6
* Fortran compiler: intel version 12.1
* architecture : amd opteron (64 bits)

* debugging : basic
* optimizations : standard

* OpenMP enabled : no (collapse: ignored)
* MPI enabled : yes
* MPI-IO enabled : auto
* GPU enabled : no (flavor: none)

* TRIO flavor = none
* TIMER flavor = abinit (libs: ignored)
* LINALG flavor = mkl (libs: user-defined)
* ALGO flavor = none (libs: ignored)
* FFT flavor = fftw3-mkl (libs: user-defined)
* MATH flavor = none (libs: ignored)
* DFT flavor = none

Then I typed :
make mj4
then I ran the parallel example that is already included in abinit using this command line:
mpirun -np 4 abinit < tbasepar_1.files>& m.log
but it did not work, pls see the attached log file.

hammouri · Post by **hammouri** » Tue Apr 07, 2015 7:36 am

Finally!!

I changed to ifort and everything went awesome. I was able to run abinit on 2, 4, 8, and 16 processors without problems.
I would like to thank you jbeuken and Jordan for your time and valuable help.

Post by **jbeuken** » Tue Apr 07, 2015 7:38 pm

As I have gfortran, I did this:
...

…
and this is the summary:

Summary of important options:

* C compiler : gnu version 4.6
* Fortran compiler: intel version 12.1
* architecture : amd opteron (64 bits)

as you can see, you do not use gfortran but ifort version 12.1 !?!

and, it's normal that there is a problem with mkl… :cry:

please, for the next time, read and re-read carefully the messages generated by the build system…
we would have gained a lot of time … :roll:

jmb

ABINIT Discussion Forums

parallel error [SOLVED]

Re: parallel error

Re: parallel error

Re: parallel error

Re: parallel error

Re: parallel error

Re: parallel error

Re: parallel error

Re: parallel error