Yambo 5.0 using Intel one API based modules

[harrier_class]

Dear developers, maintainers, and fellow users,

I worked with Yambo 5.0 with Almalinux 8 system, it worked nicely. Recently, our clusters moved to Almalinux 9 system with Intel oneAPI modules.

We have therefore moved to Yambo 5.3 for new calculations. However, we still have a few calculations remaining for an ongoing scientific paper, and we would like to complete them with Yambo 5.0 to ensure consistency with our previously obtained results.

Has anyone successfully compiled Yambo 5.0 with Intel oneAPI on AlmaLinux 9, or is there a recommended workaround or patch for this issue?

I look forward to hearing your suggestions.

--
Best regards,
Vipul

[Daniele Varsano]

Dear Vipul,

can you summarize what problems you are encountering with the installation?

Best,

Daniele

[harrier_class]

Dear Daniele,

Please find the text file with the error message and config.log attached.

As I understand the main problems during the compilation of Yambo 5.0 are arising due to the differently named compiler wrappers in the latest Intelone API, and Yambo 5.0 [being old] is not able to detect them.

PS: I am very much open to redoing the other calculations with Yambo 5.3, however, when I try to do that for the same system, however, for the same system, it shows.

Code: Select all

 <07h-59m> P106-f2169: Kernel tddft |#                                       | [002%] 04h-26m(E) 07d-08h-02m(X)

For now, I will try to increase my computation requirements and update.




Can you please help me?
--
Best regards,
Vipul

[Daniele Varsano]

Dear Vipul,

can you also share your input/report file of your calculations? Just in case input variable and parallelization strategy can be optimized.


Best,

Daniele

[Nicola Spallanzani]

Dear Vipul,
it's definitely not possible to compile Yambo 5.0 with the new Intel oneAPI compilers. You must use the older Intel compilers (icc, ifort).

Try adding these variables in the configure line:

Code: Select all

./configure CC=icc FC=ifort F77=ifort MPICC=mpiicc MPIFC=mpiifort MPIF77=mpiifort ..........

Best,
Nicola

[harrier_class]

Dear Nicola,

Thanks for your reply. Unfortunately, I dont have the older Intel compilers as the modules in my HPC cluster. I tried with the following code, with the new Intel oneAPI compilers, but it does not work [as I would expect, also I understand that your suggestion was to use these variables with the old compilers]. So, that means, no other way to compile Yambo5.0 on this new system?

Code: Select all

./configure CC=icc FC=ifort F77=ifort MPICC=mpiicc MPIFC=mpiifort MPIF77=mpiifort ..........

Dear Daniele,

I will try more, in the coming days and create new question, if needed. I should try more


I appreciate your help.

--
Best regards,
Vipul

[Nicola Spallanzani]

Dear Vipul,
don't give up, we can compile Yambo 5.0 using GCC compilers. It might be less efficient, but it definitely works.
You will need an OpenMPI module compiled with GCC.
If you don't have the hdf5, netcdf-c, and netcdf-fortran modules compiled with openmpi, don't worry, Yambo will compile them for you.

Code: Select all

./configure \
  CC=gcc FC=gfortran F77=mpifort MPICC=mpicc MPIFC=mpifort MPIF77=mpifort \
  --with-blas-libs="-L${MKLROOT}/lib -Wl,--no-as-needed -lmkl_gf_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm -ldl" \
  --with-lapack-libs="-L${MKLROOT}/lib -Wl,--no-as-needed -lmkl_gf_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm -ldl" \
  --with-scalapack-libs="-L${MKLROOT}/lib -lmkl_scalapack_lp64" \
  --with-blacs-libs="-L${MKLROOT}/lib -lmkl_blacs_openmpi_lp64" \
  --with-fft-includedir="${MKLROOT}/include" \
  --with-fft-libs="-L${MKLROOT}/lib -Wl,--no-as-needed -lmkl_gf_lp64 -lmkl_gnu_thread -lmkl_core -lgomp -lpthread -lm -ldl" \
  --enable-par-linalg --enable-mpi --enable-open-mp --enable-msgs-comps --enable-time-profile --enable-memory-profile --enable-hdf5-par-io

Let me know if this works.

Best,
Nicola

[harrier_class]

Dear Nicola,

I really appreciate your support. With your code, I managed to compile the Yambo 5.0 using gcc based compilers. Super thanks! Based on the dipole calculations, it seems that its almost 50- percent slower compared to intel based compilers. I am happy that its working, so for this project, I dont need to redo all the convergence tests


Dear Daniele,

I am happy to share that Yambo 5.3 also worked yesterday evening, as I understand the earlier implementation of TD-DFT (accessed using FxcMode ='R-def'), was super slow. New implementation works like a charm. Super thanks! For the new project, I would just go with the newest version.

Thanks a lot.

--
Best regards,
Vipul Kumar Ambasta