Skip to content

Latest commit

 

History

History

conda_env

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
README.md
README.md
 
 

README.md

How to compile Quantum ESPRESSO and Yambo in a conda environment

The following are instructions on how to prepare a conda env (python virtual environment) to then compile Quantum ESPRESSO and Yambo, and run parallel calculations.

conda create -n codes python=3.11
conda activate codes

Once activated the environment, you can proceed with the following sections. Each time you run a calculation, you need the environment to be activated, to have the correct modules...

Prepare a working openmpi env:

conda install openmpi hdf5=*=*openmpi*  fftw=*=*openmpi* libxc=*=*cpu* libnetcdf=*=*openmpi* netcdf-fortran=*=*openmpi* libblas lapack scalapack gfortran

conda install -c conda-forge ucx # to have correct MPI communication setup and run in parallel.

OR conda install yambo, to directly download the executables from the conda-forge... but I prefer to compile it by hands (see below sections).

How to install Quantum ESPRESSO

Download the Quantum ESPRESSO source from gitlab

./install/configure --with-hdf5 --enable-openmp

If you have problems with blacs, just deactivate scalapack support (as done below for kcp).

How to install Yambo

A comment on the configure: conda install the dynamic netcdf lib (you know it noticing that in the lib, it is *.so and not *.a, the latter being the static version. Yambo by default does not find the dynamical one, you have to explicitely tell him. like in the line below.)

export PATH_YAMBO_LIB="/home/jovyan/.conda/envs/codes/" # the path to your venv: check it via "conda env list"

./configure --enable-mpi --enable-open-mp --with-fft-path=$PATH_YAMBO_LIB --with-hdf5-path=$PATH_YAMBO_LIB --with-netcdf-libs=$PATH_YAMBO_LIB/lib/libnetcdf.so --with-netcdff-libs=$PATH_YAMBO_LIB/lib/libnetcdff.so --disable-hdf5-par-io --with-libxc-path=$PATH_YAMBO_LIB --with-scalapack-libs=$PATH_YAMBO_LIB/lib/libscalapack.so --with-blacs-libs=$PATH_YAMBO_LIB/lib/libscalapack.so --enable-par-linalg

Special instructions on how to run:

Follow this, if you have trouble in running in parallel:

mpirun --mca btl_vader_single_copy_mechanism none  -np 2 /home/jovyan/codes/q-e-kcw/bin/pw.x < Si.scf.in > prova.out

because:

^C^C(codes) jovyan@6f222c4867e7:~/work/koopmans_calcs/interface_yambo/trial_Si/BANDS/KI_uniq$ mpirun --mca btl_vader_single_copy_mechanism 0 -np 2 /home/jovyan/codes/q-e-kcw/bin/pw.x < Si.scf.in > prova.out
--------------------------------------------------------------------------
An invalid value was supplied for an enum variable.

  Variable     : btl_vader_single_copy_mechanism
  Value        : 0
  Valid values : 1:"cma", 4:"emulated", 3:"none"
--------------------------------------------------------------------------
--------------------------------------------------------------------------
WARNING: Linux kernel CMA support was requested via the
btl_vader_single_copy_mechanism MCA variable, but CMA support is
not available due to restrictive ptrace settings.

The vader shared memory BTL will fall back on another single-copy
mechanism if one is available. This may result in lower performance.

  Local host: 6f222c4867e7

How to compile kcp.x

Download the last version of the code. I suppose you are compiling the code in the same conda environment described above.

git clone https://github.com/epfl-theos/koopmans-kcp.git
cd koopmans-kcp
export LD_LIBRARY_PATH=/home/jovyan/.conda/envs/codes/lib # this is fundamental
./configure --with-scalapack=no MPIF90=mpif90

Before, I was launching ./configure BLAS_LIBS="-L/home/jovyan/.conda/envs/codes/lib -lopenblas" LAPACK_LIBS="-L/home/jovyan/.conda/envs/codes/lib -llapack" FFT_LIBS="-L/home/jovyan/.conda/envs/codes/lib -lfftw3" --with-scalapack=no MPIF90=mpif90, but in this way it does not put

DFLAGS         =  -D__GFORTRAN -D__FFTW3 -D__MPI -D__PARA

in the make.sys... is like it is not putting the flags because you don't allow a checking, you provide explicitely the path.

Then modify the newly generated make.sys file:

IFLAGS         = -I/home/jovyan/codes/koopmans-kcp/include -I/home/jovyan/codes/koopmans-kcp/iotk/include
MODFLAGS       = -I./  -I/home/jovyan/codes/koopmans-kcp/Modules  -I/home/jovyan/codes/koopmans-kcp/iotk/src

Otherwise it will not find the modules and iotk library. Then:

make kcp # not in parallel.

How to solve the There are not enough slots available in the system...

This happens when we run mpirun asking for more processes than the available slots. Just look for the /home/jovyan/.conda/envs/codes/etc/openmpi-default-hostfile:

find /home/jovyan/.conda/envs/codes/* -name *hostfile

and add this line:

localhost slots=24

If not enough memory... vader something

rm /dev/shm/vader*