Merge pull request #369 from trz42/nessi-2023.06-PyTorch-2.1.2-2023a-CUDA-12.1.1

{2023.06}[foss/2023a] PyTorch v2.1.2 with CUDA/12.1.1

Author: poksumdo

EESSI-install-software.sh

@@ -17,6 +17,11 @@ display_help() {
   echo "  --skip-cuda-install    -  disable installing a full CUDA SDK in the host_injections prefix (e.g. in CI)"
 }
 
+# Function to check if a command exists
+function command_exists() {
+    command -v "$1" >/dev/null 2>&1
+}
+
 function copy_build_log() {
     # copy specified build log to specified directory, with some context added
     build_log=${1}
@@ -147,6 +152,39 @@ else
   mkdir -p ${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
 fi
 
+# We need to ensure that certain files are present or updated before we source
+#   $TOPDIR/init/eessi_environment_variables
+# Particularly the files we need to have present/updated in
+#   ${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
+#   are:
+#     - .lmod/lmodrc.lua
+#     - .lmod/SitePackage.lua
+# We run scripts to create them if they don't exist or if the scripts have been
+# changed in the PR.
+
+# Set base directory for software and for Lmod config files
+_eessi_software_path=${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
+_lmod_cfg_dir=${_eessi_software_path}/.lmod
+
+# We assume there's only one diff file that corresponds to the PR patch file
+pr_diff=$(ls [0-9]*.diff | head -1)
+
+# Create or update ${_eessi_software_path}/.lmod/lmodrc.lua
+_lmodrc_file=${_lmod_cfg_dir}/lmodrc.lua
+_lmodrc_changed=$(cat ${pr_diff} | grep '^+++' | cut -f2 -d' ' | sed 's@^[a-z]/@@g' | grep '^create_lmodrc.py$' > /dev/null; echo $?)
+if [ ! -f "${_lmodrc_file}" ] || [ "${_lmodrc_changed}" == '0' ]; then
+    python3 ${TOPDIR}/create_lmodrc.py ${_eessi_software_path}
+    check_exit_code $? "${_lmodrc_file} created/updated" "Failed to create/update ${_lmodrc_file}"
+fi
+
+# Create or update ${_eessi_software_path}/.lmod/SitePackage.lua
+_lmod_sitepackage_file=${_lmod_cfg_dir}/SitePackage.lua
+_sitepackage_changed=$(cat ${pr_diff} | grep '^+++' | cut -f2 -d' ' | sed 's@^[a-z]/@@g' | grep '^create_lmodsitepackage.py$' > /dev/null; echo $?)
+if [ ! -f "${_lmod_sitepackage_file}" ] || [ "${_sitepackage_changed}" == '0' ]; then
+    python3 ${TOPDIR}/create_lmodsitepackage.py ${_eessi_software_path}
+    check_exit_code $? "${_lmod_sitepackage_file} created/updated" "Failed to create/update ${_lmod_sitepackage_file}"
+fi
+
 # Set all the EESSI environment variables (respecting $EESSI_SOFTWARE_SUBDIR_OVERRIDE)
 # $EESSI_SILENT - don't print any messages
 # $EESSI_BASIC_ENV - give a basic set of environment variables
@@ -212,13 +250,11 @@ else
     echo "Skipping installation of CUDA SDK and cu* libraries in host_injections, since the --skip-cuda-install flag was passed"
 fi
 
-# Install drivers in host_injections
-# TODO: this is commented out for now, because the script assumes that nvidia-smi is available and works;
-#       if not, an error is produced, and the bot flags the whole build as failed (even when not installing GPU software)
-# ${EESSI_PREFIX}/scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh
-
-# Don't run the Lmod GPU driver check when doing builds (may not have a GPU, and it's not relevant for vanilla builds anyway)
-export EESSI_OVERRIDE_GPU_CHECK=1
+# Install NVIDIA drivers in host_injections (if they exist)
+if command_exists "nvidia-smi"; then
+    echo "Command 'nvidia-smi' found. Installing NVIDIA drivers for use in prefix shell..."
+    ${EESSI_PREFIX}/scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh
+fi
 
 # use PR patch file to determine in which easystack files stuff was added
 changed_easystacks=$(cat ${pr_diff} | grep '^+++' | cut -f2 -d' ' | sed 's@^[a-z]/@@g' | grep '^easystacks/.*yml$' | egrep -v 'known-issues|missing') 
@@ -268,25 +304,5 @@ else
     done
 fi
 
-### add packages here
-
-echo ">> Creating/updating Lmod RC file..."
-export LMOD_CONFIG_DIR="${EASYBUILD_INSTALLPATH}/.lmod"
-lmod_rc_file="$LMOD_CONFIG_DIR/lmodrc.lua"
-lmodrc_changed=$(cat ${pr_diff} | grep '^+++' | cut -f2 -d' ' | sed 's@^[a-z]/@@g' | grep '^create_lmodrc.py$' > /dev/null; echo $?)
-if [ ! -f $lmod_rc_file ] || [ ${lmodrc_changed} == '0' ]; then
-    python3 $TOPDIR/create_lmodrc.py ${EASYBUILD_INSTALLPATH}
-    check_exit_code $? "$lmod_rc_file created" "Failed to create $lmod_rc_file"
-fi
-
-echo ">> Creating/updating Lmod SitePackage.lua ..."
-export LMOD_PACKAGE_PATH="${EASYBUILD_INSTALLPATH}/.lmod"
-lmod_sitepackage_file="$LMOD_PACKAGE_PATH/SitePackage.lua"
-sitepackage_changed=$(cat ${pr_diff} | grep '^+++' | cut -f2 -d' ' | sed 's@^[a-z]/@@g' | grep '^create_lmodsitepackage.py$' > /dev/null; echo $?)
-if [ ! -f "$lmod_sitepackage_file" ] || [ "${sitepackage_changed}" == '0' ]; then
-    python3 $TOPDIR/create_lmodsitepackage.py ${EASYBUILD_INSTALLPATH}
-    check_exit_code $? "$lmod_sitepackage_file created" "Failed to create $lmod_sitepackage_file"
-fi
-
 echo ">> Cleaning up ${TMPDIR}..."
 rm -r ${TMPDIR}

bot/build.sh

@@ -271,6 +271,11 @@ if [[ ! -z ${SHARED_FS_PATH} ]]; then
     BUILD_STEP_ARGS+=("--host-injections" "${SHARED_FS_PATH}/host-injections")
 fi
 
+# Don't run the Lmod GPU driver check when doing builds (may not have a GPU, and it's not relevant for vanilla builds anyway)
+echo "EESSI_OVERRIDE_GPU_CHECK='${EESSI_OVERRIDE_GPU_CHECK}'"
+export EESSI_OVERRIDE_GPU_CHECK=1
+echo "EESSI_OVERRIDE_GPU_CHECK='${EESSI_OVERRIDE_GPU_CHECK}'"
+
 # create tmp file for output of build step
 build_outerr=$(mktemp build.outerr.XXXX)
 

easystacks/pilot.nessi.no/2023.06/eessi-2023.06-eb-4.9.1-2023a.yml

@@ -52,3 +52,6 @@ easyconfigs:
   - ESPResSo-4.2.2-foss-2023a.eb:
       options:
         from-pr: 20595
+  - PyTorch-2.1.2-foss-2023a-CUDA-12.1.1.eb:
+      options:
+        cuda-compute-capabilities: 6.0,6.1,7.0,7.5,8.0,8.6,8.9,9.0

eb_hooks.py

@@ -425,6 +425,29 @@ def pre_configure_hook_openblas_optarch_generic(self, *args, **kwargs):
         raise EasyBuildError("OpenBLAS-specific hook triggered for non-OpenBLAS easyconfig?!")
 
 
+def pre_configure_hook_pytorch_add_cupti_libdir(self, *args, **kwargs):
+    """
+    Pre-configure hook for PyTorch: add directory $EESSI_SOFTWARE_PATH/software/CUDA/12.1.1/extras/CUPTI/lib64 to LIBRARY_PATH
+    """
+    if self.name == 'PyTorch':
+        if 'cudaver' in self.cfg.template_values and self.cfg.template_values['cudaver'] == '12.1.1':
+            _cudaver = self.cfg.template_values['cudaver']
+            print_msg("pre_configure_hook_pytorch_add_cupti_libdir: CUDA version: '%s'" % _cudaver)
+            _library_path = os.getenv('LIBRARY_PATH')
+            print_msg("pre_configure_hook_pytorch_add_cupti_libdir: library_path: '%s'", _library_path)
+            _eessi_software_path = os.getenv('EESSI_SOFTWARE_PATH')
+            print_msg("pre_configure_hook_pytorch_add_cupti_libdir: eessi_software_path: '%s'", _eessi_software_path)
+            _cupti_lib_dir = os.path.join(_eessi_software_path, 'software', 'CUDA', _cudaver, 'extras', 'CUPTI', 'lib64')
+            print_msg("pre_configure_hook_pytorch_add_cupti_libdir: cupti_lib_dir: '%s'", _cupti_lib_dir)
+            if _library_path:
+                env.setvar('LIBRARY_PATH', ':'.join([_library_path, _cupti_lib_dir]))
+            else:
+                env.setvar('LIBRARY_PATH', _cupti_lib_dir)
+            print_msg("pre_configure_hook_pytorch_add_cupti_libdir: LIBRARY_PATH: '%s'", os.getenv('LIBRARY_PATH'))
+    else:
+        raise EasyBuildError("PyTorch-specific hook triggered for non-PyTorch easyconfig?!")
+
+
 def pre_configure_hook_libfabric_disable_psm3_x86_64_generic(self, *args, **kwargs):
     """Add --disable-psm3 to libfabric configure options when building with --optarch=GENERIC on x86_64."""
     if self.name == 'libfabric':
@@ -560,6 +583,12 @@ def pre_test_hook_increase_max_failed_tests_arm_PyTorch(self, *args, **kwargs):
     """
     if self.name == 'PyTorch' and self.version == '2.1.2' and get_cpu_architecture() == AARCH64:
         self.cfg['max_failed_tests'] = 10
+        if 'cudaver' in self.cfg.template_values and self.cfg.template_values['cudaver'] == '12.1.1':
+            _cudaver = self.cfg.template_values['cudaver']
+            _runtest = self.cfg['runtest']
+            self.cfg['runtest'] = _runtest.replace(
+                                      'PYTHONUNBUFFERED',
+                                      'PYTORCH_TEST_RUN_EVERYTHING_IN_SERIAL=1 PYTHONUNBUFFERED')
 
 
 def pre_single_extension_hook(ext, *args, **kwargs):
@@ -851,6 +880,7 @@ def inject_gpu_property(ec):
     'libfabric': pre_configure_hook_libfabric_disable_psm3_x86_64_generic,
     'MetaBAT': pre_configure_hook_metabat_filtered_zlib_dep,
     'OpenBLAS': pre_configure_hook_openblas_optarch_generic,
+    'PyTorch': pre_configure_hook_pytorch_add_cupti_libdir,
     'WRF': pre_configure_hook_wrf_aarch64,
     'at-spi2-core': pre_configure_hook_atspi2core_filter_ld_library_path,
 }

eessi_container.sh

@@ -477,12 +477,6 @@ if [[ ${SETUP_NVIDIA} -eq 1 ]]; then
         mkdir -p ${EESSI_USR_LOCAL_CUDA}
         BIND_PATHS="${BIND_PATHS},${EESSI_VAR_LOG}:/var/log,${EESSI_USR_LOCAL_CUDA}:/usr/local/cuda"
         [[ ${VERBOSE} -eq 1 ]] && echo "BIND_PATHS=${BIND_PATHS}"
-        if [[ "${NVIDIA_MODE}" == "install" ]] ; then
-            # We need to "trick" our LMOD_RC file to allow us to load CUDA modules even without a CUDA driver
-            # (this works because we build within a container and the LMOD_RC recognises that)
-            touch ${EESSI_TMPDIR}/libcuda.so
-            export SINGULARITY_CONTAINLIBS="${EESSI_TMPDIR}/libcuda.so"
-        fi
     fi
 fi
 

install_scripts.sh

@@ -112,8 +112,6 @@ copy_files_by_list ${TOPDIR}/scripts ${INSTALL_PREFIX}/scripts "${script_files[@
 nvidia_files=(
     eessi-2023.06-cuda-and-libraries.yml
     install_cuda_and_libraries.sh
-    install_cuda_host_injections.sh
-    install_cuDNN_host_injections.sh
     link_nvidia_host_libraries.sh
 )
 copy_files_by_list ${TOPDIR}/scripts/gpu_support/nvidia ${INSTALL_PREFIX}/scripts/gpu_support/nvidia "${nvidia_files[@]}"

run_in_compat_layer_env.sh

@@ -29,6 +29,9 @@ fi
 if [ ! -z ${http_proxy} ]; then
     INPUT="export http_proxy=${http_proxy}; ${INPUT}"
 fi
+if [ ! -z ${EESSI_OVERRIDE_GPU_CHECK} ]; then
+    INPUT="export EESSI_OVERRIDE_GPU_CHECK=${EESSI_OVERRIDE_GPU_CHECK}; ${INPUT}"
+fi
 if [ ! -z ${https_proxy} ]; then
     INPUT="export https_proxy=${https_proxy}; ${INPUT}"
 fi