Merge pull request #269 from ocaisa/cuda_easystack_fail_2023.06

Author: bedroge

scripts/gpu_support/nvidia/easystacks/2023.06/eessi-2023.06-eb-5.3.0-2023a-CUDA-host-injections.yml

@@ -7,3 +7,5 @@ easyconfigs:
         # needed to enforce acceptance of EULA in cuDNN easyblock,
         # see https://github.com/easybuilders/easybuild-easyblocks/pull/3473
         include-easyblocks-from-commit: 11afb88ec55e0ca431cbe823696aa43e2a9bfca8
+        # Don't fail for missing PTX
+        cuda-sanity-check-accept-missing-ptx: True

scripts/gpu_support/nvidia/install_cuda_and_libraries.sh

@@ -92,7 +92,18 @@ echo "Created temporary directory '${tmpdir}'"
 # Store MODULEPATH so it can be restored at the end of each loop iteration
 SAVE_MODULEPATH=${MODULEPATH}
 
-for EASYSTACK_FILE in ${TOPDIR}/easystacks/eessi-*CUDA*.yml; do
+# Check if NVIDIA_EASYSTACKS_DIRECTORY is already set and not empty
+if [ -n "${NVIDIA_EASYSTACKS_DIRECTORY}" ]; then
+    echo_yellow "Using environment override NVIDIA_EASYSTACKS_DIRECTORY: ${NVIDIA_EASYSTACKS_DIRECTORY}"
+else
+    NVIDIA_EASYSTACKS_DIRECTORY="${TOPDIR}/easystacks"
+    echo_green "NVIDIA_EASYSTACKS_DIRECTORY environment variable override is not set, using default ${NVIDIA_EASYSTACKS_DIRECTORY}"
+fi
+# Initialize an empty array to store the processed easystack paths
+PROCESSED_EASYSTACK_FILES=()
+for EASYSTACK_FILE in ${NVIDIA_EASYSTACKS_DIRECTORY}/eessi-*CUDA*.yml; do
+    # The file must exist
+    [ -e "${EASYSTACK_FILE}" ] || continue
     echo -e "Processing easystack file ${EASYSTACK_FILE}...\n\n"
 
     # determine version of EasyBuild module to load based on EasyBuild version included in name of easystack file
@@ -267,6 +278,7 @@ for EASYSTACK_FILE in ${TOPDIR}/easystacks/eessi-*CUDA*.yml; do
       fatal_error "some installation failed, please check EasyBuild logs ${PWD}/$(basename ${eb_last_log})..."
     else
       echo_green "all installations at ${EASYBUILD_INSTALLPATH}/software/... succeeded!"
+      PROCESSED_EASYSTACK_FILES+=("${EASYSTACK_FILE}")
     fi
 
     # clean up tmpdir content
@@ -275,5 +287,16 @@ for EASYSTACK_FILE in ${TOPDIR}/easystacks/eessi-*CUDA*.yml; do
     # Restore MODULEPATH for next loop iteration
     MODULEPATH=${SAVE_MODULEPATH}
 done
+
+# Check that we processed at least one easystack file
+if [ ${#PROCESSED_EASYSTACK_FILES[@]} -gt 0 ]; then
+    echo_green "${#PROCESSED_EASYSTACK_FILES[@]} easystack files processed:"
+    for EASYSTACK_FILE in "${PROCESSED_EASYSTACK_FILES[@]}"; do
+        echo_green "- ${EASYSTACK_FILE}"
+    done
+else
+    fatal_error "Error: No matching CUDA easystack files were successfully found/processed in ${NVIDIA_EASYSTACKS_DIRECTORY}!"
+fi
+
 # Remove the temporary directory
 rm -rf "${tmpdir}"

scripts/gpu_support/nvidia/link_nvidia_host_libraries.sh

@@ -110,8 +110,8 @@ get_nvlib_list() {
 
     # see https://apptainer.org/docs/admin/1.0/configfiles.html#nvidia-gpus-cuda
     # https://github.com/apptainer/apptainer/commits/main/etc/nvliblist.conf
-    # This default_nvlib_list is based on this commit on Oct 1, 2024:
-    # https://github.com/apptainer/apptainer/commit/a19fa01527a8914839b8d1649688f83c61ba9ad2
+    # This default_nvlib_list is based on this commit on July 24, 2026:
+    # https://github.com/apptainer/apptainer/commit/b86c0aa171357c7be265d3130ab4f13a4061422f
     # TODO: driver version which corresponds to?
     local default_nvlib_list=(
         "libcuda.so"
@@ -130,6 +130,9 @@ get_nvlib_list() {
         "libglx.so"
         "libGLX.so"
         "libnvcuvid.so"
+        "libnvcucompat.so"
+        "libnvcudla.so"
+        "libnvcuextend.so"
         "libnvidia-cbl.so"
         "libnvidia-cfg.so"
         "libnvidia-compiler.so"
@@ -150,10 +153,23 @@ get_nvlib_list() {
         "libnvidia-opencl.so"
         "libnvidia-opticalflow.so"
         "libnvidia-ptxjitcompiler.so"
+        "libnvidia-rmapi-tegra.so"
         "libnvidia-rtcore.so"
         "libnvidia-tls.so"
         "libnvidia-wfb.so"
         "libnvoptix.so"
+        "libnvos.so"
+        "libnvosd.so"
+        "libnvrm_chip.so"
+        "libnvrm_gpu.so"
+        "libnvrm_host1x.so"
+        "libnvrm_mem.so"
+        "libnvrm_stream.so"
+        "libnvrm_surface.so"
+        "libnvrm_sync.so"
+        "libnvsciipc.so"
+        "libnvsocsys.so"
+        "libnvtegrahv.so"
         "libOpenCL.so"
         "libOpenGL.so"
         "libvdpau_nvidia.so"
@@ -820,4 +836,3 @@ fi
 
 # If everything went OK, show success message
 echo_green "Host NVIDIA GPU drivers linked successfully for EESSI"
-