comp-chem/cp2k
1
0
Fork 0
mirror of https://github.com/cp2k/cp2k.git synced 2026-07-29 06:35:28 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

compilation fixes (#1987)

Browse source 
* Fix compilation issue on AMD hardware

* Fix spla compilation
This commit is contained in:
Taillefumier Mathieu 2022-03-11 21:03:48 +01:00 • committed by GitHub
parent cc8ca8cdb1
commit e563fa0c91
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
2 changed files with 22 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

6
src/pw/gpu/pw_gpu_kernels_cuda.cu
View file

@ -61,7 +61,7 @@ __global__ void pw_complex_to_real(const double *zin, double *dout,
* \author Ole Schütt
******************************************************************************/
void pw_gpu_launch_complex_to_real(const double *zin, double *dout,
const int ngpts, cudaStream_t stream) {
const int ngpts, offloadStream_t stream) {
const int threadsPerBlock = 1024;
const int numBlocks = (ngpts + threadsPerBlock - 1) / threadsPerBlock;
pw_complex_to_real<<<numBlocks, threadsPerBlock, 0, stream>>>(zin, dout,
@ -87,7 +87,7 @@ __global__ void pw_gather_z(double *pwcc, const double *c, const double scale,
******************************************************************************/
void pw_gpu_launch_gather_z(double *pwcc, const double *c, const double scale,
const int ngpts, const int *ghatmap,
cudaStream_t stream) {
offloadStream_t stream) {
const int threadsPerBlock = 32;
const int numBlocks = (ngpts + threadsPerBlock - 1) / threadsPerBlock;
pw_gather_z<<<numBlocks, threadsPerBlock, 0, stream>>>(pwcc, c, scale, ngpts,
@ -118,7 +118,7 @@ __global__ void pw_scatter_z(double *c, const double *pwcc, const double scale,
******************************************************************************/
void pw_gpu_launch_scatter_z(double *c, const double *pwcc, const double scale,
const int ngpts, const int nmaps,
const int *ghatmap, cudaStream_t stream) {
const int *ghatmap, offloadStream_t stream) {
const int threadsPerBlock = 32;
const int numBlocks = (ngpts + threadsPerBlock - 1) / threadsPerBlock;
pw_scatter_z<<<numBlocks, threadsPerBlock, 0, stream>>>(c, pwcc, scale, ngpts,

27
tools/toolchain/scripts/stage8/install_spla.sh
View file

@ -94,6 +94,7 @@ case "$with_spla" in
fi
if [ "$ENABLE_HIP" = "__TRUE__" ]; then
case "${GPUVER}" in
K20X | K40 | K80 | P100 | V100 | A100)
[ -d build-cuda ] && rm -rf "build-cuda"
@ -112,9 +113,9 @@ case "$with_spla" in
${EXTRA_CMAKE_FLAGS} .. \
> cmake.log 2>&1 || tail -n ${LOG_LINES} cmake.log
make -j $(get_nprocs) > make.log 2>&1 || tail -n ${LOG_LINES} make.log
install -d ${pkg_install_dir}/lib/cuda
[ -f src/libspla.a ] && install -m 644 src/*.a ${pkg_install_dir}/lib/cuda >> install.log 2>&1
[ -f src/libspla.so ] && install -m 644 src/*.so ${pkg_install_dir}/lib/cuda >> install.log 2>&1
install -d ${pkg_install_dir}/lib/hip
[ -f src/libspla.a ] && install -m 644 src/*.a ${pkg_install_dir}/lib/hip >> install.log 2>&1
[ -f src/libspla.so ] && install -m 644 src/*.so ${pkg_install_dir}/lib/hip >> install.log 2>&1
;;
Mi50 | Mi100 | Mi200)
[ -d build-hip ] && rm -rf "build-hip"
@ -133,9 +134,9 @@ case "$with_spla" in
${EXTRA_CMAKE_FLAGS} .. \
> cmake.log 2>&1 || tail -n ${LOG_LINES} cmake.log
make -j $(get_nprocs) > make.log 2>&1 || tail -n ${LOG_LINES} make.log
install -d ${pkg_install_dir}/lib/rocm
[ -f src/libspla.a ] && install -m 644 src/*.a ${pkg_install_dir}/lib/rocm >> install.log 2>&1
[ -f src/libspla.so ] && install -m 644 src/*.so ${pkg_install_dir}/lib/rocm >> install.log 2>&1
install -d ${pkg_install_dir}/lib/hip
[ -f src/libspla.a ] && install -m 644 src/*.a ${pkg_install_dir}/lib/hip >> install.log 2>&1
[ -f src/libspla.so ] && install -m 644 src/*.so ${pkg_install_dir}/lib/hip >> install.log 2>&1
;;
*) ;;
esac
@ -151,7 +152,7 @@ case "$with_spla" in
SPLA_CFLAGS="-I'${pkg_install_dir}/include/spla'"
SPLA_LDFLAGS="-L'${pkg_install_dir}/lib' -Wl,-rpath='${pkg_install_dir}/lib'"
SPLA_CUDA_LDFLAGS="-L'${pkg_install_dir}/lib/cuda' -Wl,-rpath='${pkg_install_dir}/lib/cuda'"
SPLA_HIP_LDFLAGS="-L'${pkg_install_dir}/lib/rocm' -Wl,-rpath='${pkg_install_dir}/lib/rocm'"
SPLA_HIP_LDFLAGS="-L'${pkg_install_dir}/lib/hip' -Wl,-rpath='${pkg_install_dir}/lib/hip'"
;;
__SYSTEM__)
@ -195,15 +196,25 @@ EOF
export SPLA_CFLAGS="${SPLA_CFLAGS}"
export SPLA_LDFLAGS="${SPLA_LDFLAGS}"
export SPLA_CUDA_LDFLAGS="${SPLA_CUDA_LDFLAGS}"
export SPLA_HIP_LDFLAGS="${SPLA_HIP_LDFLAGS}"
export CP_DFLAGS="\${CP_DFLAGS} IF_MPI(-D__SPLA|)"
export CP_CFLAGS="\${CP_CFLAGS} ${SPLA_CFLAGS}"
export CP_LDFLAGS="\${CP_LDFLAGS} IF_CUDA(${SPLA_CUDA_LDFLAGS}|${SPLA_LDFLAGS})"
export SPLA_LIBRARY="-lspla"
export SPLA_ROOT="$pkg_install_dir"
export SPLA_INCLUDE_DIR="$pkg_install_dir/include/spla"
export SPLA_VERSION=${spla-ver}
export CP_LIBS="IF_MPI(${SPLA_LIBS}|) \${CP_LIBS}"
EOF
if [ "$ENABLE_HIP" = "__TRUE__" ]; then
cat << EOF >> "${BUILDDIR}/setup_spla"
export CP_LDFLAGS="\${CP_LDFLAGS} IF_HIP(${SPLA_HIP_LDFLAGS}|${SPLA_LDFLAGS})"
EOF
fi
if [ "$ENABLE_CUDA" = "__TRUE__" ]; then
cat << EOF >> "${BUILDDIR}/setup_spla"
export CP_LDFLAGS="\${CP_LDFLAGS} IF_CUDA(${SPLA_CUDA_LDFLAGS}|${SPLA_LDFLAGS})"
EOF
fi
cat "${BUILDDIR}/setup_spla" >> $SETUPFILE
fi