CP2K 6.1.0 for LX
ウェブページ
https://www.cp2k.org/バージョン
6.1.0ビルド環境
- Intel Parallel Studio 2017 Update 8
- CUDA Toolkit 9.1.85 (only for GPU version)
- spglib/1.11.1
ビルドに必要なファイル
- cp2k-6.1.0.tar.gz
- plumed-2.4.3.tgz
- (以下のスクリプト内でもいくつかファイルを入手しています)
- install_openmpi.patch
--- scripts/install_openmpi.sh.org 2018-11-20 10:08:10.000000000 +0900
+++ scripts/install_openmpi.sh 2018-11-20 10:07:47.000000000 +0900
@@ -56,13 +56,13 @@
;;
__SYSTEM__)
echo "==================== Finding OpenMPI from system paths ==========
=========="
- check_command mpirun "openmpi"
- check_command mpicc "openmpi"
- check_command mpif90 "openmpi"
- check_command mpic++ "openmpi"
- check_lib -lmpi "openmpi"
- add_include_from_paths OPENMPI_CFLAGS "mpi.h" $INCLUDE_PATHS
- add_lib_from_paths OPENMPI_LDFLAGS "libmpi.*" $LIB_PATHS
+ #check_command mpirun "openmpi"
+ #check_command mpicc "openmpi"
+ #check_command mpif90 "openmpi"
+ #check_command mpic++ "openmpi"
+ #check_lib -lmpi "openmpi"
+ #add_include_from_paths OPENMPI_CFLAGS "mpi.h" $INCLUDE_PATHS
+ #add_lib_from_paths OPENMPI_LDFLAGS "libmpi.*" $LIB_PATHS
;;
__DONTUSE__)
;;
@@ -90,24 +90,26 @@
else
mpi_bin=mpirun
fi
- # check openmpi version as reported by mpirun
- raw_version=$($mpi_bin --version 2>&1 | \
- grep "(Open MPI)" | awk '{print $4}')
- major_version=$(echo $raw_version | cut -d '.' -f 1)
- minor_version=$(echo $raw_version | cut -d '.' -f 2)
- # old versions required -lmpi_cxx to link cxx code, new version don't
- if [ $major_version -gt 1 ] ; then
+ ## check openmpi version as reported by mpirun
+ #raw_version=$($mpi_bin --version 2>&1 | \
+ # grep "(Open MPI)" | awk '{print $4}')
+ #major_version=$(echo $raw_version | cut -d '.' -f 1)
+ #minor_version=$(echo $raw_version | cut -d '.' -f 2)
+ ## old versions required -lmpi_cxx to link cxx code, new version don't
+ #if [ $major_version -gt 1 ] ; then
OPENMPI_LIBS="-lmpi"
- else
- OPENMPI_LIBS="-lmpi -lmpi_cxx"
- fi
+ #else
+ # OPENMPI_LIBS="-lmpi -lmpi_cxx"
+ #fi
# old versions didn't support MPI 3, so adjust __MPI_VERSION accordingly (needed e.g. for pexsi)
- if [ $major_version -lt 1 ] || \
- [ $major_version -eq 1 -a $minor_version -lt 7 ] ; then
- mpi2_dflags="-D__MPI_VERSION=2"
- else
+ #if [ $major_version -lt 1 ] || \
+ # [ $major_version -eq 1 -a $minor_version -lt 7 ] ; then
+ # mpi2_dflags="-D__MPI_VERSION=2"
+ #else
mpi2_dflags=''
- fi
+ #fi
+ OPENMPI_CFLAGS="-I${I_MPI_ROOT}/include64"
+ OPENMPI_LDFLAGS="-L${I_MPI_ROOT}/lib64 -Wl,-rpath=${I_MPI_ROOT}/lib64"
cat <<EOF >> "${BUILDDIR}/setup_openmpi"
export OPENMPI_CFLAGS="${OPENMPI_CFLAGS}"
export OPENMPI_LDFLAGS="${OPENMPI_LDFLAGS}"
- install_mkl_intel_intelmpi.patch
--- scripts/install_mkl.sh.org 2018-11-12 14:44:45.000000000 +0900
+++ scripts/install_mkl.sh 2018-11-12 14:50:40.000000000 +0900
@@ -66,7 +66,7 @@
fi
done
# set the correct lib flags from MLK link adviser
- MKL_LIBS="-Wl,--start-group ${mkl_lib_dir}/libmkl_gf_lp64.a ${mkl_lib_dir}/
libmkl_core.a ${mkl_lib_dir}/libmkl_sequential.a"
+ MKL_LIBS="-lmkl_intel_lp64 -lmkl_sequential -lmkl_core"
# check optional libraries
if [ $MPI_MODE != no ] ; then
enable_mkl_scalapack="__TRUE__"
@@ -74,11 +74,11 @@
case $MPI_MODE in
mpich)
mkl_optional_libs="$mkl_optional_libs libmkl_blacs_lp64.a"
- mkl_blacs_lib="libmkl_blacs_lp64.a"
+ mkl_blacs_lib="-lmkl_blacs_lp64"
;;
openmpi)
mkl_optional_libs="$mkl_optional_libs libmkl_blacs_openmpi_lp64
.a"
- mkl_blacs_lib="libmkl_blacs_openmpi_lp64.a"
+ mkl_blacs_lib="-lmkl_blacs_intelmpi_lp64"
;;
*)
enable_mkl_scalapack="__FALSE__"
@@ -91,13 +91,13 @@
done
if [ $enable_mkl_scalapack = "__TRUE__" ] ; then
echo "Using MKL provided ScaLAPACK and BLACS"
- MKL_LIBS="${mkl_lib_dir}/libmkl_scalapack_lp64.a ${MKL_LIBS} ${mkl_lib_dir}/${mkl_blacs_lib}"
+ MKL_LIBS="-lmkl_scalapack_lp64 ${MKL_LIBS} ${mkl_blacs_lib}"
fi
else
enable_mkl_scalapack="__FALSE__"
fi
- MKL_LIBS="${MKL_LIBS} -Wl,--end-group -lpthread -lm -ldl"
- MKL_CFLAGS="${MKL_CFLAGS} -I${MKLROOT}/include"
+ MKL_LIBS="-L${mkl_lib_dir}/lib/intel64 ${MKL_LIBS} -lpthread -lm"
+ MKL_CFLAGS="${MKL_CFLAGS} -I${MKLROOT}/include"
# write setup files
cat <<EOF > "${BUILDDIR}/setup_mkl"
- install_elpa.patch
--- scripts/install_elpa.sh.org 2018-11-16 16:22:07.000000000 +0900
+++ scripts/install_elpa.sh 2018-11-16 16:29:20.000000000 +0900
@@ -92,6 +92,8 @@
LDFLAGS="-Wl,--enable-new-dtags ${MATH_LDFLAGS} ${SCALAPACK_LDFLAGS} ${cray_ldflags}" \
LIBS="${SCALAPACK_LIBS} ${MATH_LIBS}" \
> configure.log 2>&1
+ sed -i -e "s/^wl=\"\"/wl=\"-Wl,\"/" \
+ -e "s/^pic_flag=\"\"/pic_flag=\" -fPIC\"/" libtool
make -j $NPROCS > make.log 2>&1
make install > install.log 2>&1
cd ..
@@ -114,6 +116,8 @@
LDFLAGS="-Wl,--enable-new-dtags ${MATH_LDFLAGS} ${SCALAPACK_LDFLAGS} ${cray_ldflags}" \
LIBS="${SCALAPACK_LIBS} ${MATH_LIBS}" \
> configure.log 2>&1
+ sed -i -e "s/^wl=\"\"/wl=\"-Wl,\"/" \
+ -e "s/^pic_flag=\"\"/pic_flag=\" -fPIC\"/" libtool
make -j $NPROCS > make.log 2>&1
make install > install.log 2>&1
cd ..
- install_quip.patch
--- scripts/install_quip.sh.org 2018-11-16 16:39:15.000000000 +0900
+++ scripts/install_quip.sh 2018-11-16 16:50:31.000000000 +0900
@@ -66,34 +66,24 @@
-e "s|\(cd build/.*\)|\1 >&- 2>&-|g" \
bin/find_sizeof_fortran_t
fi
- sed -i \
- -e "s|\(F77 *=\).*|\1 ${FC}|g" \
- -e "s|\(F90 *=\).*|\1 ${FC}|g" \
- -e "s|\(F95 *=\).*|\1 ${FC}|g" \
- -e "s|\(CC *=\).*|\1 ${CC}|g" \
- -e "s|\(CPLUSPLUS *=\).*|\1 ${CXX}|g" \
- -e "s|\(LINKER *=\).*|\1 ${FC}|g" \
- -e "s|\(FPP *=\).*|\1 ${FC} -E -x f95-cpp-input|g" \
- -e "s|\(QUIPPY_FCOMPILER *=\).*|\1 ${FC}|g" \
- -e "s|\(QUIPPY_CPP *=\).*|\1 ${FC} -E -x f95-cpp-input|g" \
- arch/Makefile.linux_${quip_arch}_gfortran
# enable debug symbols
- echo "F95FLAGS += -g" >> arch/Makefile.linux_${quip_arch}_gfo
rtran
- echo "F77FLAGS += -g" >> arch/Makefile.linux_${quip_arch}_gfo
rtran
- echo "CFLAGS += -g" >> arch/Makefile.linux_${quip_arch}_gfo
rtran
- echo "CPLUSPLUSFLAGS += -g" >> arch/Makefile.linux_${quip_arch}_gfortran
- export QUIP_ARCH=linux_${quip_arch}_gfortran
+ echo "F95FLAGS += -g" >> arch/Makefile.linux_${quip_arch}_ifort_icc
+ echo "F77FLAGS += -g" >> arch/Makefile.linux_${quip_arch}_ifort_icc
+ echo "CFLAGS += -g" >> arch/Makefile.linux_${quip_arch}_ifort_icc
+ echo "CPLUSPLUSFLAGS += -g" >> arch/Makefile.linux_${quip_arch}_ifort_icc
+ export QUIP_ARCH=linux_${quip_arch}_ifort_icc
# hit enter a few times to accept defaults
- echo -e "${MATH_LDFLAGS} ${MATH_LIBS} \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n" | make config > configure.log
+ QUIP_MATH_LIBS="-lmkl_intel_lp64 -lmkl_sequential -lmkl_core -lpthread -lm -ldl"
+ echo -e "${MATH_LDFLAGS} ${QUIP_MATH_LIBS} \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n" | make config > configure.log
# make -j does not work :-(
make > make.log 2>&1
! [ -d "${pkg_install_dir}/include" ] && mkdir -p "${pkg_install_dir}/include"
! [ -d "${pkg_install_dir}/lib" ] && mkdir -p "${pkg_install_dir}/lib"
- cp build/linux_x86_64_gfortran/quip_unified_wrapper_module.mod \
+ cp build/linux_x86_64_ifort_icc/quip_unified_wrapper_module.mod \
"${pkg_install_dir}/include/"
- cp build/linux_x86_64_gfortran/*.a \
+ cp build/linux_x86_64_ifort_icc/*.a \
"${pkg_install_dir}/lib/"
- cp src/FoX-4.0.3/objs.linux_${quip_arch}_gfortran/lib/*.a \
+ cp src/FoX-4.0.3/objs.linux_${quip_arch}_ifort_icc/lib/*.a \
"${pkg_install_dir}/lib/"
cd ..
touch "${install_lock_file}"
ビルド手順
#!/bin/sh
INSTDIR=/local/apl/lx/cp2k610
WORKDIR=/work/users/${USER}
VERSION=6.1.0
TARBALL=/home/users/${USER}/Software/CP2K/${VERSION}/cp2k-${VERSION}.tar.gz
#LIBGRID=/home/users/${USER}/Software/CP2K/${VERSION}/libgrid/cp2k-${VERSION}-i1708/tools/autotune_grid
PATCH_INST_OPENMPI=/home/users/${USER}/Software/CP2K/${VERSION}/install_openmpi.patch
PATCH_INST_MKL=/home/users/${USER}/Software/CP2K/${VERSION}/install_mkl_intel_intelmpi.patch
PATCH_INST_ELPA=/home/users/${USER}/Software/CP2K/${VERSION}/install_elpa.patch
PATCH_INST_QUIP=/home/users/${USER}/Software/CP2K/${VERSION}/install_quip.patch
PLUMED_VERSION=2.4.3
PLUMED_TARBALL=/home/users/${USER}/Software/PLUMED/${PLUMED_VERSION}/plumed-${PLUMED_VERSION}.tgz
PARALLEL=12
# NOTE: Some of header and library paths are defined in CPATH and
# LIBRARY_PATH environment variables from modulefiles.
#---------------------------------------------------------------------------
umask 0022
export LANG=""
export LC_ALL=C
module purge
module load intel_parallelstudio/2017update8
module load spglib/1.11.1
cd $INSTDIR
if [ -d cp2k-${VERSION} ]; then
mv cp2k-${VERSION} cp2k-erase
rm -rf cp2k-erase &
fi
tar zxf ${TARBALL}
mv cp2k-${VERSION}/* .
rm -rf cp2k-${VERSION}/{.dockerignore,.gitignore}
rmdir cp2k-${VERSION}
#rm -rf tools/autotune_grid
#cp -rf ${LIBGRID} tools/autotune_grid
#cp ${LIBGRID}/libgrid.a tools/autotune_grid
# plumed
mkdir ${INSTDIR}/plumed
cd $WORKDIR
if [ -d plumed-${PLUMED_VERSION} ]; then
mv plumed-${PLUMED_VERSION} plumed-erase
rm -rf plumed-erase &
fi
tar zxf ${PLUMED_TARBALL}
cd plumed-${PLUMED_VERSION}
CC=mpiicc FC=mpiifort CXX=mpiicpc \
./configure --prefix=${INSTDIR}/plumed
make -j ${PARALLEL}
make check
make install
# cp2k toolchain
cd ${INSTDIR}/tools/toolchain
sed -e "s/blacs_openmpi/blacs_intelmpi/" scripts/install_mkl.sh > scripts/install_mkl.sh.impi
chmod 755 scripts/install_mkl.sh.impi
patch -p0 < $PATCH_INST_OPENMPI
patch -p0 < $PATCH_INST_MKL
patch -p0 < $PATCH_INST_ELPA
patch -p0 < $PATCH_INST_QUIP
CC=icc FC=ifort F77=ifort F90=ifort CXX=icpc \
MPICC=mpiicc MPICXX=mpiicpc MPIFC=mpiifort MPIF77=mpiifort MPIF90=mpiifort \
./install_cp2k_toolchain.sh --math-mode=mkl \
--mpi-mode=openmpi \
--with-elpa=install \
--with-cmake=system \
--with-mpich=no \
--with-openmpi=system \
--with-reflapack=no \
-j ${PARALLEL}
# cheat the script
mv -f scripts/install_mkl.sh.impi scripts/install_mkl.sh
CC=icc FC=ifort F77=ifort F90=ifort CXX=icpc \
MPICC=mpiicc MPICXX=mpiicpc MPIFC=mpiifort MPIF77=mpiifort MPIF90=mpiifort \
./install_cp2k_toolchain.sh --math-mode=mkl \
--mpi-mode=openmpi \
--with-elpa=install \
--with-ptscotch=install \
--with-parmetis=install \
--with-superlu=install \
--with-pexsi=install \
--with-quip=install \
--with-cmake=system \
--with-mpich=no \
--with-openmpi=system \
--with-reflapack=no \
-j ${PARALLEL}
#LIBGRID_ESC=`echo ${INSTDIR}/tools/autotune_grid | sed -e 's/\//\\\\\//g'`
# finish building toolchain, copy arch file to the proper place
## modify arch for cpu-only version
sed -e "/^CFLAGS =/s/CFLAGS =/CFLAGS = -O2 /" \
-e "/^FCFLAGS =/s/FCFLAGS =/FCFLAGS = -O2 /" \
-e "/^DFLAGS /s/\$/ -D__SPGLIB -D__PLUMED2/" \
-e "/^LIBS /s/\$/ -lsymspg -lz -lgsl -nofor_main/" \
install/arch/local.psmp > ${INSTDIR}/arch/rccs.psmp
echo "include ${INSTDIR}/plumed/lib/plumed/src/lib/Plumed.inc" >> ${INSTDIR}/arch/rccs.psmp
echo "EXTERNAL_OBJECTS=\$(PLUMED_STATIC_DEPENDENCIES)" >> ${INSTDIR}/arch/rccs.psmp
## modify arch for cpu-gpu version
sed -e "/^CFLAGS =/s/CFLAGS =/CFLAGS = -O2 /" \
-e "/^FCFLAGS =/s/FCFLAGS =/FCFLAGS = -O2 /" \
-e "/^DFLAGS /s/\$/ -D__SPGLIB -D__PLUMED2 -D__ACC -D__DBCSR_ACC -D__PW_CUDA/" \
-e "/^LIBS /s/\$/ -lsymspg -lz -lgsl -lcudart -lcufft -lcublas -nofor_main/" \
install/arch/local.psmp > ${INSTDIR}/arch/rccs_cuda.psmp
echo "include ${INSTDIR}/plumed/lib/plumed/src/lib/Plumed.inc" >> ${INSTDIR}/arch/rccs_cuda.psmp
echo "EXTERNAL_OBJECTS=\$(PLUMED_STATIC_DEPENDENCIES)" >> ${INSTDIR}/arch/rccs_cuda.psmp
echo "NVCC = nvcc -D__GNUC__=5 -D__GNUC_MINOR__=3 -Xcompiler=--std=gnu++98" >> ${INSTDIR}/arch/rccs_cuda.psmp
echo "NVFLAGS = -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 \$(DFLAGS)" >> ${INSTDIR}/arch/rccs_cuda.psmp
#--
cd ${INSTDIR}/makefiles
make -j ${PARALLEL} ARCH=rccs VERSION=psmp
module load cuda/9.1
make -j ${PARALLEL} ARCH=rccs_cuda VERSION=psmp
テスト
plumed
インストールのログより。3つのエラーは全て軽微な数値エラー+ check file ves/rt-td-vonmises/report.txt for more information
+ ERROR in test isdb/rt-emmi/
+ check file isdb/rt-emmi/report.txt for more information
+ ERROR in test isdb/rt-jcouplings-mi/
+ check file isdb/rt-jcouplings-mi/report.txt for more information
+ ERROR in test isdb/rt-jcouplings/
+ check file isdb/rt-jcouplings/report.txt for more information
+++++++++++++++++++++++++++++++++++++++++++++++++++++
+ Final report:
+ 248 tests performed, 161 tests not applicable
+ 3 errors found
+ Find the bug!
+ To replace references, go to the test directory and
+ type 'make reset'
+++++++++++++++++++++++++++++++++++++++++++++++++++++
cp2k
テスト結果は /local/apl/lx/cp2k610/regtesting/rccs/psmp もしくは /local/apl/lx/cp2k610/regtesting/rccs_cuda/psmp に残しています。GPU版以外は ccfep 上でテストを実行しています。
#!/bin/sh
export LC_ALL=C
export LANG=""
module purge
module load intel_parallelstudio/2017update8
module load spglib/1.11.1
CP2K=/local/apl/lx/cp2k610/
CP2K_ARCH=rccs
CP2K_VER=psmp
TIMEOUT=120
PARALLEL=16
ulimit -s unlimited
cd ${CP2K}/regtesting/${CP2K_ARCH}/${CP2K_VER}
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
# serial test
../../../tools/regtesting/do_regtest \
-nobuild \
-nosvn \
-arch ${CP2K_ARCH} \
-version ${CP2K_VER} \
-mpiranks 1 \
-ompthreads 1 \
-jobmaxtime ${TIMEOUT} \
-cp2kdir ../../../ \
-maxtasks ${PARALLEL} >& regtest_mpi1_omp1.log
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
# omp test
../../../tools/regtesting/do_regtest \
-nobuild \
-nosvn \
-arch ${CP2K_ARCH} \
-version ${CP2K_VER} \
-mpiranks 1 \
-ompthreads 2 \
-jobmaxtime ${TIMEOUT} \
-cp2kdir ../../../ \
-maxtasks ${PARALLEL} >& regtest_mpi1_omp2.log
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
# mpi test
../../../tools/regtesting/do_regtest \
-nobuild \
-nosvn \
-arch ${CP2K_ARCH} \
-version ${CP2K_VER} \
-mpiranks 2 \
-ompthreads 1 \
-jobmaxtime ${TIMEOUT} \
-cp2kdir ../../../ \
-maxtasks ${PARALLEL} >& regtest_mpi2_omp1.log
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
# mpi/openmp test
../../../tools/regtesting/do_regtest \
-nobuild \
-nosvn \
-arch ${CP2K_ARCH} \
-version ${CP2K_VER} \
-mpiranks 2 \
-ompthreads 2 \
-jobmaxtime ${TIMEOUT} \
-cp2kdir ../../../ \
-maxtasks ${PARALLEL} >& regtest_mpi2_omp2.log
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
# yet another mpi test
../../../tools/regtesting/do_regtest \
-nobuild \
-nosvn \
-arch ${CP2K_ARCH} \
-version ${CP2K_VER} \
-mpiranks 8 \
-ompthreads 1 \
-jobmaxtime ${TIMEOUT} \
-cp2kdir ../../../ \
-maxtasks ${PARALLEL} >& regtest_mpi8_omp1.log
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
# yet another mpi/openmp test
../../../tools/regtesting/do_regtest \
-nobuild \
-nosvn \
-arch ${CP2K_ARCH} \
-version ${CP2K_VER} \
-mpiranks 8 \
-ompthreads 2 \
-jobmaxtime ${TIMEOUT} \
-cp2kdir ../../../ \
-maxtasks ${PARALLEL} >& regtest_mpi8_omp2.log
rm -rf LAST-${CP2K_ARCH}-${CP2K_VER}
テスト結果: MPI1 - OMP1
--------------------------------- Summary --------------------------------
Number of FAILED tests 1
Number of WRONG tests 0
Number of CORRECT tests 3008
Number of NEW tests 14
Total number of tests 3023
GREPME 1 0 3008 14 3023 X
--------------------------------------------------------------------------
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
テスト結果: MPI1 - OMP2
--------------------------------- Summary --------------------------------
Number of FAILED tests 1
Number of WRONG tests 0
Number of CORRECT tests 3008
Number of NEW tests 14
Total number of tests 3023
GREPME 1 0 3008 14 3023 X
--------------------------------------------------------------------------
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
テスト結果: MPI2 - OMP1
--------------------------------- Summary --------------------------------
Number of FAILED tests 1
Number of WRONG tests 0
Number of CORRECT tests 3053
Number of NEW tests 19
Total number of tests 3073
GREPME 1 0 3053 19 3073 X
--------------------------------------------------------------------------
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
テスト結果: MPI2 - OMP2
--------------------------------- Summary --------------------------------
Number of FAILED tests 1
Number of WRONG tests 0
Number of CORRECT tests 3053
Number of NEW tests 19
Total number of tests 3073
GREPME 1 0 3053 19 3073 X
--------------------------------------------------------------------------
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
テスト結果: MPI8 - OMP1
--------------------------------- Summary --------------------------------
Number of FAILED tests 8
Number of WRONG tests 2
Number of CORRECT tests 3012
Number of NEW tests 16
Total number of tests 3038
GREPME 8 2 3012 16 3038 X
--------------------------------------------------------------------------
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
- QS/regtest-rma-3D/* : RUNTIME FAIL
テスト結果: MPI8 - OMP2
--------------------------------- Summary --------------------------------
Number of FAILED tests 8
Number of WRONG tests 2
Number of CORRECT tests 3012
Number of NEW tests 16
Total number of tests 3038
GREPME 8 2 3012 16 3038 X
--------------------------------------------------------------------------
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
- QS/regtest-rma-3D/* : RUNTIME FAIL
テスト結果: GPU版 MPI2 - OMP2 (@ccca)
--------------------------------- Summary --------------------------------
Number of FAILED tests 3
Number of WRONG tests 0
Number of CORRECT tests 3052
Number of NEW tests 19
Total number of tests 3074
GREPME 3 0 3052 19 3074 X
--------------------------------------------------------------------------
- QS/regtest-pao-2/H2O_pao_rotinv.inp : RUNTIME FAIL
- QS/regtest-rel/Hg_rel.inp : RUNTIME FAIL
- QS/regtest-ri-mp2/opt_basis_O_auto_gen.inp : RUNTIME FAIL
ベンチマーク
tests/QS/benchmark/H2O-64.inp を利用。(時間は grep "CP2K " *.log で表示される値から)| jobtype | 総コア数 (ノード数) |
MPI | OMP | GPU | elapse(sec) |
| core | 18 (1) | 18 | 1 | - | 77.148 |
| small | 40 (1) | 40 | 1 | - | 45.537 |
| small | 80 (2) | 80 | 1 | - | 36.421 |
| small | 160 (4) | 160 | 1 | - | 27.490 |
| small | 160 (4) | 80 | 2 | - | 33.158 |
| small | 160 (4) | 32 | 5 | - | 27.090 |
| gpu | 12 (1) | 12 | 1 | 1 | 130.869 |
雑多な情報
- 公式サイトのコンパイラ対応情報 https://www.cp2k.org/dev:compiler_support
- libsmm は未検証(x86_64 では xsmm で代用可能と判断)
- intel17-intelmpi とは異なり、intel17-openmpi3.1.0 では少しエラーが増える(mpi1*omp1の条件でも)
- 全体的に openmpi よりは intelmpi の方が正常に動作するように見える
- intel17-openmpi3.1.0-mkl2017.x の条件ではなぜか MPI 並列が異様に遅くなる
- intel17-openmpi3.1.0-mkl2018.x ではなぜか MPI 並列の問題が解消する
- 今回のビルド条件では libgrid.a が OpenMP 並列時に正常に動作せず(ビルド時の -qopenmp の有無に関わらず)
- intel17-openmpi3.1.0-mkl の場合ならば -qopenmp 付きでビルドした libgrid.a が正常に動作
- libgrid.a は正常に動作する場合には 5% 程度の速度アップが望める
- intel18 ではエラーが少し増えるため今回は回避
- intel mpiの他バージョンについては未検証
- -O3 -xHost -ip の最適化条件で H2O-64 のベンチマークをおこなったところ、-O2 だけの方が高速だった
- GPU 版はかなり大きな系(>1000原子)でないと効率的ではないらしい