Initial archdetect support for RISC-V and custom init script for RISC-V repo

.github/workflows/scorecards.yml

@@ -40,7 +40,7 @@ jobs:
           persist-credentials: false
 
       - name: "Run analysis"
-        uses: ossf/scorecard-action@99c53751e09b9529366343771cc321ec74e9bd3d # v2.0.6
+        uses: ossf/scorecard-action@0864cf19026789058feabb7e87baa5f140aac736 # v2.3.1
         with:
           results_file: results.sarif
           results_format: sarif

.github/workflows/tests_archdetect.yml

@@ -13,6 +13,8 @@ jobs:
           - x86_64/intel/skylake_avx512/archspec-linux-6132
           - x86_64/amd/zen2/Azure-CentOS7-7V12
           - x86_64/amd/zen3/Azure-CentOS7-7V73X
+          - x86_64/amd/zen4/Azure-Alma8-9V33X
+          - x86_64/amd/zen4/Shinx-RHEL8-9654
           - aarch64/neoverse_n1/Azure-Ubuntu20-Altra
           - aarch64/neoverse_n1/AWS-awslinux-graviton2
           - aarch64/neoverse_v1/AWS-awslinux-graviton3

.github/workflows/tests_scripts.yml

@@ -3,9 +3,9 @@ name: Tests for scripts
 on:
   push:
     paths:
-      - build_container.sh
       - create_directory_tarballs.sh
       - create_lmodsitepackage.py
+      - eessi_container.sh
       - EESSI-install-software.sh
       - install_software_layer.sh
       - load_easybuild_module.sh
@@ -15,9 +15,9 @@ on:
 
   pull_request:
     paths:
-      - build_container.sh
       - create_directory_tarballs.sh
       - create_lmodsitepackage.py
+      - eessi_container.sh
       - EESSI-install-software.sh
       - install_software_layer.sh
       - load_easybuild_module.sh

EESSI-install-software.sh

@@ -147,6 +147,24 @@ else
   mkdir -p ${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
 fi
 
+# if we run the script for the first time, e.g., to start building for a new
+#   stack, we need to ensure certain files are present in
+#   ${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
+# - .lmod/lmodrc.lua
+# - .lmod/SitePackage.lua
+_eessi_software_path=${EESSI_PREFIX}/software/${EESSI_OS_TYPE}/${EESSI_SOFTWARE_SUBDIR_OVERRIDE}
+_lmod_cfg_dir=${_eessi_software_path}/.lmod
+_lmod_rc_file=${_lmod_cfg_dir}/lmodrc.lua
+if [ ! -f ${_lmod_rc_file} ]; then
+    command -V python3
+    python3 ${TOPDIR}/create_lmodrc.py ${_eessi_software_path}
+fi
+_lmod_sitepackage_file=${_lmod_cfg_dir}/SitePackage.lua
+if [ ! -f ${_lmod_sitepackage_file} ]; then
+    command -V python3
+    python3 ${TOPDIR}/create_lmodsitepackage.py ${_eessi_software_path}
+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
@@ -203,10 +221,21 @@ ${TOPDIR}/install_scripts.sh --prefix ${EESSI_PREFIX}
 # Hardcode this for now, see if it works
 # TODO: We should make a nice yaml and loop over all CUDA versions in that yaml to figure out what to install
 # Allow skipping CUDA SDK install in e.g. CI environments
+# The install_cuda... script uses EasyBuild. So, we need to check if we have EB
+# or skip this step.
+module_avail_out=$TMPDIR/ml.out
+module avail 2>&1 | grep EasyBuild &> ${module_avail_out}
+if [[ $? -eq 0 ]]; then
+    echo_green ">> Found an EasyBuild module"
+else
+    echo_yellow ">> No EasyBuild module found: skipping step to install CUDA (see output in ${module_avail_out})"
+    export skip_cuda_install=True
+fi
+
 if [ -z "${skip_cuda_install}" ] || [ ! "${skip_cuda_install}" ]; then
     ${EESSI_PREFIX}/scripts/gpu_support/nvidia/install_cuda_host_injections.sh -c 12.1.1 --accept-cuda-eula
 else
-    echo "Skipping installation of CUDA SDK in host_injections, since the --skip-cuda-install flag was passed"
+    echo "Skipping installation of CUDA SDK in host_injections, since the --skip-cuda-install flag was passed OR no EasyBuild module was found"
 fi
 
 # Install drivers in host_injections

README.md

@@ -9,14 +9,17 @@ See also https://www.eessi.io/docs/software_layer .
 You can set up your environment by sourcing the init script:
 
 ```
-$ source /cvmfs/software.eessi.io/versions/2023.06/init/bash
-Found EESSI repo @ /cvmfs/software.eessi.io/versions/2023.06!
-Derived subdirectory for software layer: x86_64/intel/haswell
-Using x86_64/intel/haswell subdirectory for software layer
+$ source /cvmfs/riscv.eessi.io/versions/20240402/init/bash
+Found EESSI repo @ /cvmfs/riscv.eessi.io/versions/20240402!
+archdetect says riscv64/generic
+Using riscv64/generic as software subdirectory.
+Found Lmod configuration file at /cvmfs/riscv.eessi.io/versions/20240402/software/linux/riscv64/generic/.lmod/lmodrc.lua
+Found Lmod SitePackage.lua file at /cvmfs/riscv.eessi.io/versions/20240402/software/linux/riscv64/generic/.lmod/SitePackage.lua
+Using /cvmfs/riscv.eessi.io/versions/20240402/software/linux/riscv64/generic/modules/all as the directory to be added to MODULEPATH.
 Initializing Lmod...
-Prepending /cvmfs/software.eessi.io/versions/2023.06/software/x86_64/intel/haswell/modules/all to $MODULEPATH...
-Environment set up to use EESSI (2023.06), have fun!
-[EESSI 2023.06] $
+Prepending /cvmfs/riscv.eessi.io/versions/20240402/software/linux/riscv64/generic/modules/all to $MODULEPATH...
+Environment set up to use EESSI (20240402), have fun!
+{EESSI 20240402} $
 ```
 
 ### Accessing EESSI via a container

create_lmodsitepackage.py

@@ -20,6 +20,28 @@
     return content
 end
 
+local function from_eessi_prefix(t)
+    -- eessi_prefix is the prefix with official EESSI modules
+    -- e.g. /cvmfs/software.eessi.io/versions/2023.06
+    local eessi_prefix = os.getenv("EESSI_PREFIX")
+
+    -- If EESSI_PREFIX wasn't defined, we cannot check if this module was from the EESSI environment
+    -- In that case, we assume it isn't, otherwise EESSI_PREFIX would (probably) have been set
+    if eessi_prefix == nil then
+        return False
+    else
+        -- NOTE: exact paths for site so may need to be updated later.
+        -- See https://github.com/EESSI/software-layer/pull/371
+
+        -- eessi_prefix_host_injections is the prefix with site-extensions (i.e. additional modules)
+        -- to the official EESSI modules, e.g. /cvmfs/software.eessi.io/host_injections/2023.06
+        local eessi_prefix_host_injections = string.gsub(eessi_prefix, 'versions', 'host_injections')
+
+       -- Check if the full modulepath starts with the eessi_prefix_*
+        return string.find(t.fn, "^" .. eessi_prefix) ~= nil or string.find(t.fn, "^" .. eessi_prefix_host_injections) ~= nil
+    end
+end
+
 local function load_site_specific_hooks()
     -- This function will be run after the EESSI hooks are registered
     -- It will load a local SitePackage.lua that is architecture independent (if it exists) from e.g.
@@ -153,10 +175,13 @@
 -- Combine both functions into a single one, as we can only register one function as load hook in lmod
 -- Also: make it non-local, so it can be imported and extended by other lmodrc files if needed
 function eessi_load_hook(t)
-    eessi_cuda_enabled_load_hook(t)
+    -- Only apply CUDA hooks if the loaded module is in the EESSI prefix
+    -- This avoids getting an Lmod Error when trying to load a CUDA module from a local software stack
+    if from_eessi_prefix(t) then
+        eessi_cuda_enabled_load_hook(t)
+    end
 end
 
-
 hook.register("load", eessi_load_hook)
 
 -- Note that this needs to happen at the end, so that any EESSI specific hooks can be overwritten by the site