ABINIT Discussion Forums

Hello everyone：
I find a strange phenomenon:
With the same compiling parameters, the abinit 8.8.1 is much slower than abinit 8.6.3.
e.g for Al with vacumn slabs, the abinit 8.6.3 is about ~6 min, but the abinit 8.8.1 needs near 30+ min.
Why?
PS: virtual machine, 4 cores, 8 GB, CentOS 7.3 with openmpi paralel.
Thanks a lot！

Dear Jianz,
Are you really using the OpenMP parallelism or just MPI? If you are just running MPI, I advise to switch off the OpenMP.
Please also have a look what is discussed here about OMP_NUM_THREADS, etc:
https://forum.abinit.org/viewtopic.php?f=3&t=3829
Best wishes,
Eric

Dear Eric:

Thanks for your answer.
My meaning is, with the same compiling parameters, and the same running command, the New version is much slower than the old ones.
Usually the two version should give the nearly same running result, I think.

I also find it is the same fast for very small computation, but the difference is large for a little heavy computation.

I wonder if the newer version change some development structure ?
hence the compiling parameters is OK for the old one, but not for the newer one ?
Maybe before I find out the reason, I have to return back the old version.

Thanks again.

Dear Jianz,
Could you send your config.log file such as we can have a look and see what can be the problem?
Best wishes,
Eric

Dear Eric,
Thanks first for you kind help.
My main compiling parameters are as following:
./configure FC=mpif90 CC=mpicc CXX=mpicxx --prefix="/opt/ABINIT-Simple" --with-arch=x86_64 --enable-64bit-flags="yes" --enable-mpi="yes" --with-mpi-libs="-L/opt/MPICH3/lib -lmpi" --with-mpi-incs="-I/opt/MPICH3/include" --enable-mpi-io="yes" --enable-openmp="yes" --with-openmp-prefix="/opt/OPENMP3/bin" --with-openmp-libs="-L/opt/OPENMPI3/lib -lmpi" --with-openmp-incs="-I/opt/OPENMPI3/include" --enable-fftw3="yes" --with-fft-incs="-I/opt/FFTW3/include/" --with-fft-libs="-L/opt/FFTW3/lib -lfftw3_omp -lfftw3_mpi -lfftw3" --enable-libxc="yes" --with-libxc-incs="-I/opt/LIBXC4/include" --with-libxc-libs="-L/opt/LIBXC4/lib -lxcf03 -lxcf90 -lxc" --enable-atompaw="yes" --with-atompaw-bins="/opt/ATOMPAW4/bin" --with-atompaw-libs="-L/opt/ATOMPAW4/lib -latompaw" --enable-gw-dpc="yes"
The compiling is OK for both ABINIT 8.6.3 and 8.8.1.
The test input file is Al-04.in.

For ABINIT 8.6.3, it needs 6'08'', but for ABINIT 8.8.1, the running needs over 30+ min.

All the works are under VMWare Machinies, with CentOS 7.3, 4 cores, 8 GB.

I have change some compiling parameters, but the running results are the same.
So I don't know why!

Thanks.

Dear Jianz,
I've made a quick test on my machine (Linux Ubuntu 2*6 cores Xeon, gfortran compilation) and I have more or less the same calculation time on the input you sent for 8.8.1 and 8.6.3.
So far I don't see what can be your problem, could you re-test by enforcing export OMP_NUM_THREADS=1 to check if this can be the source of the problem?
Best wishes,
Eric

Thanks Eric.
I have reinstall my virtual machine with the newest CentOS-1804 and re-compiling my ABINIT-8.6.3 and ABINIT-8.8.1 under Windows 10 system.
This time, I don't use the shared folder. All files are within the virtual machine.
With the same input file, the ABINIT 8.6.3 takes 6'26'', but the 8.8.1 takes 10'22'' .
To my knowledge, the only perhaps reason is the virtual machine???
Next time, I will install these two editions on the real CentOS 7.4 system and compare them.
OK, till then I will post the result.
If the result is the same, maybe I have to wait for the next new edition.