<div dir="ltr"><div>Hi,</div><div>I compiled a CUDA compatible version of CP2K, but when I submit my jobs, I get the following error:</div><div>--------------------------------------------------------------------------<br>ERROR: cudaErrorStubLibrary /scratch/ag9288/softwares/cp2k-2022.2/src/offload/offload_library.c 49<br>forrtl: error (76): Abort trap signal<br>Image              PC                Routine            Line        Source             <br>cp2k.psmp          0000000006B83CAB  Unknown               Unknown  Unknown<br>libpthread-2.28.s  000014D73245EB20  Unknown               Unknown  Unknown<br><a href="http://libc-2.28.so">libc-2.28.so</a>       000014D712D4037F  gsignal               Unknown  Unknown<br><a href="http://libc-2.28.so">libc-2.28.so</a>       000014D712D2ADB5  abort                 Unknown  Unknown<br>cp2k.psmp          00000000031F8C29  offload_get_devic          49  offload_library.c<br>cp2k.psmp          0000000001387F46  f77_interface_mp_         276  f77_interface.F<br>cp2k.psmp          000000000042281D  MAIN__                    284  cp2k.F<br>cp2k.psmp          0000000006BED096  Unknown               Unknown  Unknown<br><a href="http://libc-2.28.so">libc-2.28.so</a>       000014D712D2C493  __libc_start_main     Unknown  Unknown<br>cp2k.psmp          000000000042195E  Unknown               Unknown  Unknown<br>--------------------------------------------------------------------------</div><div><br></div><div>It looks like the executable is unable to find the card, while all the libraries are correctly loaded.</div><div><br></div><div>Version: <br></div><div>Intel fortran: 19.1</div><div>cuda : 11.3</div><div>openmpi 4.1.1</div><div>CP2K: 2022.2</div><div><br></div><div>Steps I took to build the binary:</div><div>1. Install the toolchain:</div><div>./install_cp2k_toolchain.sh
 --with-intel=system --enable-cuda --gpu-ver=V100  --with-openmpi=system
  --with-libxc=install --with-libint=install --with-fftw=install 
--with-mkl=system --with-libxsmm=install  --with-elpa=install   
--with-ptscotch=no   --with-superlu=no    --with-pexsi=no     
 --with-quip=no       --with-plumed=no    --with-sirius=no     
--with-gsl=no        --with-libvdwxc=no   --with-spglib=no   
 --with-hdf5=no    --with-spfft=no      --with-spla=install   
--with-cosma=install  --with-libvori=no --with-openblas=no -j4</div><div><br></div><div>2. Manualy build and copy Libint, for the fortran makefile bug:</div><div><a href="https://github.com/cp2k/libint-cp2k/releases">https://github.com/cp2k/libint-cp2k/releases</a><br>cmake
 -DCMAKE_C_COMPILER=icc -DCMAKE_C_FLAGS="-qopenmp -O2" 
-DCMAKE_CXX_COMPILER=icpc -DCMAKE_CXX_FLAGS="-qopenmp -O2" 
-DCMAKE_Fortran_COMPILER=ifort -DCMAKE_Fortran_FLAGS="-qopenmp -O2" 
-DREQUIRE_CXX_API=OFF -DENABLE_FORTRAN=ON ..</div><div><br></div><div>3. Apply required CUDA patch to CP2K and compile:</div><div><a href="https://github.com/cp2k/cp2k/commit/ee6c3aa.patch">https://github.com/cp2k/cp2k/commit/ee6c3aa.patch</a></div><div> make -j 4 ARCH=local_cuda VERSION="ssmp sdbg psmp pdbg"</div><div><br></div><div>PS:
 The job-node is different from the compile node, could that might be an
 issue? Also while the compilation went fine, there was an error in 
symlinking the binaries.<br></div></div>

<p></p>

-- <br />
You received this message because you are subscribed to the Google Groups "cp2k" group.<br />
To unsubscribe from this group and stop receiving emails from it, send an email to <a href="mailto:cp2k+unsubscribe@googlegroups.com">cp2k+unsubscribe@googlegroups.com</a>.<br />
To view this discussion on the web visit <a href="https://groups.google.com/d/msgid/cp2k/CA%2Bbx8_z%2Bc6FRbQF0a%2BxDyO4pmJq4%3DiKf5Y0_yLApGkNaov_77g%40mail.gmail.com?utm_medium=email&utm_source=footer">https://groups.google.com/d/msgid/cp2k/CA%2Bbx8_z%2Bc6FRbQF0a%2BxDyO4pmJq4%3DiKf5Y0_yLApGkNaov_77g%40mail.gmail.com</a>.<br />