[CP2K-user] [CP2K:20565] Re: Segmentation Fault in MKL

[Frederick Stein]

Dear Jack,
There is either bug in the libraries or your combination of OpenMPI/MKL 
does not work together. Have you ever used this combination of 
MKL/OpenMPI/GCC elsewhere? Can you check your MPI compilers and your 
mpiexec/mpirun commands?
Best,
Frederick

Jack Draney schrieb am Mittwoch, 14. August 2024 um 05:27:08 UTC+2:

> Hi all,
>
> I'm trying to build CP2K with cmake, and I'm running into some trouble. 
> The compile and install of DBCSR and CP2K go fine, but often calculations 
> will segfault in the middle of execution. This happens 
> in  cp_fm_diag::cp_fm_syevd_base (cp_fm_diag.F:609) then in MKL (see image 
> for full stacktrace)
> [image: Screenshot 2024-08-13 at 4.43.00 PM.png]
> Reason: address not mapped to object (attempt to access invalid address)
>
> This occurs with MKL versions 2020.1, 2022.2.0, and 2024.0 as well as CP2K 
> version 2023.21, 2024.1, and 2024.2. I have tried many things and must be 
> missing something someone better equipped might be able to help me with. 
> I've removed all dependencies but dbcsr to help isolate the issue.
> Here is my installation script for the two packages on Princeton's Della 
> cluster:
>
> module purge
> module load cudatoolkit/11.7
> module load gcc/8
> module load openmpi/gcc/4.1.0
> module load fftw/gcc/openmpi-4.1.0/3.3.9
> module load gsl/2.6
> module load hdf5/gcc/openmpi-4.1.0/1.10.6
> module load intel-mkl/2020.1
>
> CP2K_DIR=/scratch/gpfs/DGRAVES/cp2k-2
> INSTALL_DIR=$CP2K_DIR/install
> export CMAKE_PREFIX_PATH=$CMAKE_PREFIX_PATH:$INSTALL_DIR
>
> cd $CP2K_DIR
>
> FYPP_VERSION=3.2
> wget https://github.com/aradi/fypp/archive/refs/tags/${FYPP_VERSION}
> .tar.gz
> tar -xzf ${FYPP_VERSION}.tar.gz
> mv fypp-${FYPP_VERSION} fypp
> export PATH=$PATH:$CP2K_DIR/fypp/bin
>
> DBCSR_VERSION=2.7.0
> wget https://github.com/cp2k/dbcsr/archive/refs/tags/v${DBCSR_VERSION}
> .tar.gz
> tar -xzf v${DBCSR_VERSION}.tar.gz
> mv dbcsr-${DBCSR_VERSION} dbcsr
> cd dbcsr
> mkdir build
> cd build
> cmake \
> -DUSE_MPI=ON \
> -DUSE_OPENMP=ON \
> -DUSE_ACCEL=cuda \
> -DWITH_GPU=A100 \
> -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR \
> ..
> make -j 16
> make install
>
> cd $CP2K_DIR
>
> CP2K_VERSION=2024.2
> git clone -b support/v${CP2K_VERSION} --recursive 
> https://github.com/cp2k/cp2k.git cp2k
> cd cp2k
>
> cmake -S . -B build \
> -DCMAKE_BUILD_TYPE=Release \
> -DCP2K_USE_SIRIUS=OFF \
> -DCP2K_USE_LIBINT2=OFF \
> -DCP2K_USE_LIBXC=OFF \
> -DCP2K_USE_SPGLIB=OFF \
> -DCP2K_USE_ELPA=OFF \
> -DCP2K_USE_SPLA=OFF \
> -DCP2K_USE_ACCEL=CUDA \
> -DCP2K_WITH_GPU=A100 \
> -DCP2K_BLAS_VENDOR=MKL \
> -DCP2K_SCALAPACK_VENDOR=MKL \
> -DCMAKE_INSTALL_PREFIX=$INSTALL_DIR \
> .
>
>
>
> If anyone can provide any guidance, I would be greatly appreciative.
>
> Best regards,
> Jack Draney
>

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/375db0dc-8407-4256-b0cd-164c54ce3a5bn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240814/6bc37fac/attachment-0001.htm>