EPW benchmarking
Posted: Sat Jul 22, 2023 9:30 am
Hello,
I'm trying to reproduce the results reported for the recently reported code benchmarking using MgB2 on OLCF Summit ((https://docs.epw-code.org/doc/Benchmarks.html) in order to optimize my current build of the code. Could you please share some insight into how EPW v6.0alpha was compiled there (e.g., was it compiled with gcc? which libraries/modules?)
So far, I have compiled the code with the following:
]The code seems to run smoothly for the most part (especially for the small test examples), however, for MgB2, it seg faults ("Program received signal SIGSEGV: Segmentation fault - invalid memory reference") and dumps cores shortly after it finishes calculating all the irreducible q-points and correctly writing the .epb files (see the attached file for a calculation that was parallelized over $NPROC=$NPOOL for both the nsfc and epw parts).
I have also attempted to compile with the NVHPC instead, but the error persists.
Thank you!
Kindly,
Matthew
I'm trying to reproduce the results reported for the recently reported code benchmarking using MgB2 on OLCF Summit ((https://docs.epw-code.org/doc/Benchmarks.html) in order to optimize my current build of the code. Could you please share some insight into how EPW v6.0alpha was compiled there (e.g., was it compiled with gcc? which libraries/modules?)
So far, I have compiled the code with the following:
Code: Select all
Currently Loaded Modules:
1) gcc/9.1.0 2) spectrum-mpi/10.4.0.3-20210112 3) hdf5/1.10.7 4) netlib-lapack/3.9.1 5) essl/6.3.0
export BLAS_LIBS="-L$OLCF_ESSL_ROOT/lib64 -lessl"
export LAPACK_LIBS="-L$OLCF_ESSL_ROOT/lib64 -lessl $OLCF_NETLIB_LAPACK_ROOT/lib64/liblapack.a"
export F90=gfortran CC=gcc CXX=g++
./configure --enable-openmp --with-hdf5=$OLCF_HDF5_ROOT
sed -i "/DFLAGS/s/__FFTW/__LINUX_ESSL/" make.inc
sed -i "/CFLAGS/s/= /= -c11 /" make.inc
make -j32 allI have also attempted to compile with the NVHPC instead, but the error persists.
Thank you!
Kindly,
Matthew