[QE-users] use of blas&lapack from amd-optimized libflame 2.1 causes test failures in epw tests

Evgeny Permyakov permeakra at gmail.com
Sun Feb 2 22:23:14 CET 2020


on ubuntu 18.04 / epyc 7302 machine with stock gfortran and openmpi I
compiled libflame 2.1 from source with config string "
--enable-builtin-blas' '--enable-lapack2flame'
'--enable-vector-intrinsics=sse' '--enable-memory-alignment=8'
'--enable-ldim-alignment' '--prefix=/home/permeakra/comp-soft/prefix/'
'--enable-shared ", built and install.
I also installed amd-optimized fftw into the same prefix, configured with
'--enable-sse2' '--enable-avx' '--enable-avx2' '--enable-mpi'
'--enable-openmp' '--enable-shared' '--enable-amd-opt'
'--prefix=/home/permeakra/comp-soft/prefix'

Then I took recent quantum espresso and linked to those libraries for fftw,
blas and lapack. This, in addition to ./config with line
'BLAS_LIBS=-lflame' 'LAPACK_LIBS=-lflame -llapacke' 'FFT_LIBS=-lfftw3'
required manual setting of LD_LIBS  in generated make.inc to
-L/home/permeakra/comp-soft/prefix/lib . Then I compiled by running **make
all && make eps && make links**, entered test-suite  subdirectory and run
parallel teststuite with **make run-tests-parallel** . majority of the
tests passed except for some **epw** tests with rather large absolute error.
(sidenote: I had to set  **orte_execute_quiet = true** in
~/.openmpi/mca-params.conf so mpi didn't mess with extra messages

Naturally, I checked again with  amd-optimized fftw, but internal blas &
lapack coming with QE. There was no errors in epw tests.

I don't use epw in my work and pwscf tests work just fine, but the fact
that some tests failed bother me. Should I investigate it further and how?
I'm not versed in QE code, I'm a humble user.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20200203/d4f975b4/attachment.html>


More information about the users mailing list