Try to change the subdir in which the CUDA toolkit is installed so that it also doesnt include the CPU microarchitecture

Author: Caspar van Leeuwen

easystacks/software.eessi.io/2023.06/accel/nvidia/rebuilds/20250805-eb-5.1.1-rebuild-2023a-for-cuda-sanity-check.yml

@@ -0,0 +1,10 @@
+# We'll build all CUDA software, for various reasons
+# 1. We now have a proper CUDA sanity check, and if anything was 'wrong' with our current CUDA installs, we'd like
+# to know about it
+# 2. The PR implementing a CI to check for differences between officially supported CUDA Compute Capabilities shows
+# that there are a lot of missing installations https://github.com/EESSI/software-layer/pull/1087 . A rebuild PR like
+# this will have the convenient side effect of filling all those holes
+easyconfigs:
+  - CUDA-12.1.1.eb:
+      options:
+        accept-eula-for: CUDA

eb_hooks.py

@@ -151,7 +151,7 @@ def parse_list_of_dicts_env(var_name):
     if not re.match(r'^[A-Za-z_][A-Za-z0-9_]*$', var_name):
         raise ValueError(f"Invalid environment variable name: {var_name}")
     list_string = os.getenv(var_name, '[]')
-    
+
     list_of_dicts = []
     try:
         # Try JSON format first
@@ -162,7 +162,7 @@ def parse_list_of_dicts_env(var_name):
             list_of_dicts = ast.literal_eval(list_string)
         except (ValueError, SyntaxError):
             raise ValueError(f"Environment variable '{var_name}' does not contain a valid list of dictionaries.")
-    
+
     return list_of_dicts
 
 
@@ -211,7 +211,7 @@ def post_ready_hook(self, *args, **kwargs):
         parallel = self.parallel
     else:
         parallel = self.cfg['parallel']
-    
+
     if parallel == 1:
         return  # no need to limit if already using 1 core
 
@@ -733,7 +733,7 @@ def pre_configure_hook_score_p(self, *args, **kwargs):
 def pre_configure_hook_vsearch(self, *args, **kwargs):
     """
     Pre-configure hook for VSEARCH
-    - Workaround for a Zlib macro being renamed in Gentoo, see https://bugs.gentoo.org/383179 
+    - Workaround for a Zlib macro being renamed in Gentoo, see https://bugs.gentoo.org/383179
       (solves "expected initializer before 'OF'" errors)
     """
     if self.name == 'VSEARCH':
@@ -1301,13 +1301,28 @@ def replace_non_distributable_files_with_symlinks(log, install_dir, pkg_name, al
                     log.debug("%s is not found in allowlist, so replacing it with symlink: %s",
                               print_name, full_path)
                     # the host_injections path is under a fixed repo/location for CUDA or cuDNN
+                    # full_path is something similar to
+                    # /cvmfs/software.eessi.io/version/.../x86_64/amd/zen4/accel/nvidia/cc90/.../CUDA/bin/nvcc
+                    # host_inj_path will then be
+                    # /cvmfs/software.eessi.io/host_injections/.../x86_64/amd/zen4/accel/nvidia/cc90/.../CUDA/bin/nvcc
                     host_inj_path = re.sub(EESSI_INSTALLATION_REGEX, HOST_INJECTIONS_LOCATION, full_path)
                     # CUDA and cu* libraries themselves don't care about compute capability so remove this
                     # duplication from under host_injections (symlink to a single CUDA or cu* library
                     # installation for all compute capabilities)
                     accel_subdir = get_eessi_envvar("EESSI_ACCELERATOR_TARGET")
+                    software_subdir = get_eessi_envvar("EESSI_SOFTWARE_SUBDIR")
+                    cpu_family = get_eessi_envvar("EESSI_CPU_FAMILY")
+                    # If accel_subdir is defined, remove it from the full path
+                    # After removal of accel_subdir, host_inj_path will be something like
+                    # /cvmfs/software.eessi.io/host_injections/.../x86_64/amd/zen4/.../CUDA/bin/nvcc
                     if accel_subdir:
-                        host_inj_path = host_inj_path.replace("/accel/%s" % accel_subdir, '')
+                        host_inj_path = host_inj_path.replace(accel_subdir, '')
+                    # /cvmfs/software.eessi.io/host_injections/.../x86_64/amd/zen4/.../CUDA/bin/nvcc
+                    # If software_subdir is defined (it should always be...), replace it by only the cpu_family
+                    # After this substitution, host_inj_path will be something like
+                    # /cvmfs/software.eessi.io/host_injections/.../x86_64/.../CUDA/bin/nvcc
+                    if software_subdir and cpu_family:
+                        host_inj_path.replace(software_subdir, cpu_family)
                     # make sure source and target of symlink are not the same
                     if full_path == host_inj_path:
                         raise EasyBuildError("Source (%s) and target (%s) are the same location, are you sure you "

scripts/gpu_support/nvidia/install_cuda_and_libraries.sh

@@ -132,8 +132,15 @@ for EASYSTACK_FILE in ${TOPDIR}/easystacks/eessi-*CUDA*.yml; do
 
     # If there is a GPU on the node, the installation path will by default have an
     # accelerator subdirectory. For CUDA and cu*, these are binary installations and
-    # don't care about the target compute capability. Our hooks are aware of this and
-    # therefore expect CUDA to be available under EESSI_SITE_SOFTWARE_PATH
+    # don't care about the target compute capability nor the CPU microarchitecture.
+    # Our hooks are aware of this and therefore expect CUDA to be available under 
+    # something like EESSI_SITE_SOFTWARE_PATH, but then with the CPU micro-architecture
+    # stripped
+    # This sed command will capture everything from the EESSI_SITE_SOFTWARE_PATH up until
+    # the EESSI_SOFTWARE_SUBDIR in a capture group. It will the replace that with the content
+    # of the capture group and then have the EESSI_CPU_FAMILY appended
+    # Thus EESSI_SITE_CPU_FAMILY_PATH is something like /cvmfs/software.eessi.io/host_injections/.../x86_64
+    EESSI_SITE_CPU_FAMILY_PATH=$(echo "$EESSI_SITE_SOFTWARE_PATH" | sed 's/\(.*\)'"$EESSI_SOFTWARE_SUBDIR"'/\1'"$EESSI_CPU_FAMILY"'/')
     export EASYBUILD_INSTALLPATH=$EESSI_SITE_SOFTWARE_PATH
 
     # Install modules in hidden .modules dir to keep track of what was installed before