--enable-par-linalg but SERIAL linear algebra at runtime

[mrefiore]

Dear Yambo developers and users,

I have recently compiled Yambo on a local cluster specifying --enable-par-linalg in the configure command. I didn't specify any external mathematical library and the report I got accordingly shows that they were compiled in their internal version:

# > MATH: (Internal FFTW3)
#
# [ Ic] FFT : /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/libfftw3.a
# -I/home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/include/
# [ Ic] BLAS : /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/libblas.a
# [ Ic] LAPACK : /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/liblapack.a
# [ Ic] SCALAPACK : /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/libscalapack.a
# [ Ic] BLACS : /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/libblacs.a /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/libblacs_C_init.a /home/mrefiorentin/yamboA/yambo-4.4.0/lib/external/gfortran/mpifort/lib/libblacs_init.a
# [ - ] PETSC :
#
# [ - ] SLEPC :

However, when I run any calculation I get

P0001: [LA] SERIAL linear algebra

Is it consistent?
I expect that PARALLEL linear algebra would speed up the calculations. What should I specify in configure to run Yambo with parallel linear algebra?
Thank you very much for your attention!
All the best


Michele Re Fiorentin

[Daniele Varsano]

Dear Michele,
in order to use parallel linear algebra (SCALAPACK) you need to have it compiled and it seems to be the case and assign tasks (e.g. a number of cpu different from 1) to the scalapack operation in the input file e.g.:

Code: Select all

BS_nCPU_LinAlg_INV
BS_nCPU_LinAlg_DIAGO

for BSE
or

Code: Select all

X_nCPU_LinAlg_INV

for response function.
These variables appear by adding the -V par verbosity option (or -V all).
Note that even in the case SERIAL linalg is used for diagonalization or inversion of matrices, this does not means that the job is serial.
Best,
Daniele

[mrefiore]

Dear Daniele,

Thank you very much for your quick reply! I didn't switch on parallelization on LA in the input files.
All the best!


Michele