Yambo Community Forum

Dear all,

I am compiling YAMBO 4.0 on the Swiss HPC facilitty Piz Daint.
I had few questions I would like to be answered on how to maximise the code.
I know it could be system dependent and I will make some testing when more expert, in the meantime I would like to have some advice.

The 1st one is related to Scalapack and BLACS, according to the configure file they are not present. No anymore at least since in the previous version (and the example on the website) they were. Is this true?
If this is the case why they have been removed?

The 2nd question is related to mpi-libraries, What would they be? Can they really speed-up a code?

The last question is related to which libraries pass to the compiler. Is this a reasonable choice?

--with-fft-libs="-L$MKL_LIBRARY -lmkl_cdft_core"
--with-blas-libs="-L$MKL_LIBRARY -lmkl_sequential -lmkl_core"
--with-lapack-libs="-L$MKL_LIBRARY -lmkl_intel_lp64 -lmkl_core"

or is it better using the FFTW3?

Cheers
Davide

Dear Davide,

1. Scalapack and BLACS.
They were present before, but not effective. Now we are working on that, and we are in testing phase. A release with a full scalapack support will be released soon.

2. MPI libraries: I cannot advise you on that as I did not do estensive benchmark tests, I do not know which libraries you have available, and should be all equivalent.

3. MKL for sure is a good choice.

May some other else have tested different options and can give you more advises.

Best,
Daniele

Dear Davide,

just to follow up on Daniele's answer.

* full support to BLACS and Scalapack will be released with yambo 4.1 (beta release already available, full release due in few days).
* on intel machines, if you have intel compilers and mkl, I would go for them... Even better if you have mpiifort (though we do not have
massive numerical evidence)
* the mil choice you are reporting is totally ok
concerning FFT, I would use those internal to mkl (though I personally do not expect massive changes wrt FFTW3)
if you want you can also get black and scalapack from mkl

take care
Andrea