How to install Yambo on a GPU cluster

[jasonhan0710]

Dear all,

I have noticed that yambo 5.0.x have extended CUDA support. However I have no idea to install it on a GPU machine. Would you please tell me how to install it with CUDA? Thank you!

Best,
Jason

[Daniele Varsano]

Dear Jason,

Yambo is ported using CUDA Fortran, you need to compile Yambo using the PGI fortran compiler.
In the GPU supercomputer I use I have this settings:

Code: Select all

module load cuda/10.2
module load hpc-sdk/2020--binary
module load spectrum_mpi/10.3.1--binary

Code: Select all

export FC=pgf90
export F77=pgfortran
export CPP='cpp -E'
export CC=pgcc
export FPP="pgfortran -Mpreprocess -E"
export F90SUFFIX=".f90"
export MPIFC=mpipgifort
export MPIF77=mpipgifort
export MPICC=mpipgicc

Code: Select all

./configure FC=$FC F77=$F77 CPP="cpp -E"  CC=$CC MPIFC=$MPIFC MPIF77=$MPIF77 MPICC=$MPICC \
    --with-blas-libs="-lblas" \
    --with-lapack-libs="-llapack" \
    --enable-cuda=cuda10.2,cc70 \
    --enable-mpi --enable-open-mp \
    --enable-msgs-comps \
    --enable-time-profile \
    --enable-memory-profile

Essentially you need to use the directive

Code: Select all

--enable-cuda=cuda10.2,cc70

according to your GPU cards and compile the code with PGI compiler.

Best,

Daniele

[Dean]

Dear all，

I want to build yambo on GPU nodes. However, I just got a serial version, which shows low efficiency.
Here is my setting:

Code: Select all

export PATH=/home/dingym/bin/nvhpc_sdk/Linux_x86_64/23.3/compilers/bin:$PATH
export MANPATH=/home/dingym/bin/nvhpc_sdk/Linux_x86_64/23.3/compilers/man:$MANPATH

Code: Select all

./configure FC=nvfortran F77=nvfortran  CC=nvc CPP="cpp -E -P" FPP="pgfortran -Mpreprocess -E" --enable-cuda=cuda11.0,cc80 --enable-msgs-comps --enable-memory-profile --enable-mpi

The setup file is attached.
Any help is appreciated!Thanks in advance.

[Daniele Varsano]

Dear Yimin,
it seems that MPI library are not linked. The config.log file would help to spot the reason this happened.
You can try to assign the MPI from the configure as:
./configure CC=nvc FC=nvfortran FPP="nvfortran -Mpreprocess -E" MPIFC=mpif90 ....

and see if it is recognized.
Best,
Daniele

[vvu]

Dears developers!

I want to install Yambo in the IRENE machine in order to use CUDA's GPUs.
I saw in this tutorial for CPUs in IRENE:
https://www.yambo-code.eu/wiki/index.ph ... iot_Curie).
I tried to add: --enable-cuda=cuda11.0,cc70 at the end of the ./configure command for CPUS in IRENE (as in the link above) but it doesn't work.

Could you please help me?

[Daniele Varsano]

Dear vvu,
please, sign your post with your name and affiliation, this is a rule of the forum. You can do once for all by filling the signature in your user profile.

In order to understand what is failing can you please post the config.log file?

Best,

Daniele