[WIP] Update and sandardize implementation of packages, in sync with …

…spack update (#593)

* Update RSC with rocm 6.1.2 and more + reactivate rocm job for testing

* Update Radiuss Shared CI too (minor changes there)

* Fix regression in submodules

* Allow failure for new xl compiler on lassen, same error as with previous version

* From RSC: Upgrade to rocm@6.2.0 and cray-mpich@8.1.30 in CI on tioga

* Clean CI config

* From RSC: Enforce coherency between rocm software stack and compiler

* RSC Fix

* From RSC: Fix typo: rocm compiler is rocmcc

* Remove XL jobs (RSC and local), revert to cce 17.0.1 on tioga

* Point at main branch in RSC

* Update Spack

* Minor changes to build_and_test script

* Restore appropriate version of RSC

* Revert "Restore appropriate version of RSC"

This reverts commit 8fba6f2.

* Update Spack (and RSC accordingly)

* Update and standardize RADIUSS packages

* build_and_test.sh: allow to control whether or not to push to registry

* From RSC: packages fix

* From RSC: point at main branch

* Fix spack branch name

* Update Spack ref (waiting for next snapshot tag)

* Update RSC to main

* Update RSC to main

* Update Spack to appropriate snapshot branch

.gitlab/jobs/lassen.yml

@@ -21,13 +21,7 @@ variables:
 # We keep ${PROJECT_<MACHINE>_VARIANTS} and ${PROJECT_<MACHINE>_DEPS} So that
 # the comparison with the original job is easier.
 
-# Allow failure due to weird error only with XL on Lassen. Passes in debug mode.
-xl_2022_08_19_gcc_8_3_1_cuda_11_2_0:
-  variables:
-    SPEC: "${PROJECT_LASSEN_VARIANTS} +cuda %xl@=16.1.1.12.gcc.8.3.1 ^cuda@11.2.0+allow-unsupported-compilers ${PROJECT_LASSEN_DEPS}"
-    MODULE_LIST: "cuda/11.2.0"
-  extends: .job_on_lassen
-  allow_failure: true
+# No overridden jobs so far.
 
 # Allow failure due to weird error only with XL on Lassen. Passes in debug mode.
 xl_2023_06_28_gcc_11_2_1_cuda_11_8_0:

.gitlab/jobs/tioga.yml

@@ -14,14 +14,38 @@
 variables:
   INCLUDED_FILE_CANNOT_BE_EMPTY: "True"
 
+########################
+# Overridden shared jobs
+########################
+# We duplicate the shared jobs description and add necessary changes for this
+# project. We keep ${PROJECT_<MACHINE>_VARIANTS} and ${PROJECT_<MACHINE>_DEPS}
+# So that the comparison with the original job is easier.
+
+# This compiler is not considered reliable per issue in RAJA
+# https://github.com/LLNL/RAJA/pull/1712#issuecomment-2316335119
+# We allow failure and keep the cce 17.0.1 job below.
+cce_18_0_0:
+  variables:
+    SPEC: "${PROJECT_TIOGA_VARIANTS} %cce@=18.0.0 ${PROJECT_TIOGA_DEPS}"
+  extends: .job_on_tioga
+  allow_failure: true
+
+############
+# Extra jobs
+############
+# We do not recommend using ${PROJECT_<MACHINE>_VARIANTS} and
+# ${PROJECT_<MACHINE>_DEPS} in the extra jobs. There is no reason not to fully
+# describe the spec here.
 # The correct way is now to use rocmcc compiler directly as done in shared spec
-### Template for extra jobs defined by the project.
-#clang13.0.1-rocm5.4.1:
-#  variables:
-#    SPEC: "%clang@13.0.1 +rocm amdgpu_target=gfx90a ^hip@5.4.1"
-#  extends: .job_on_tioga
 
+# As long as cce 18 fails, we keep this one.
+cce_17_0_1:
+  variables:
+    SPEC: "${PROJECT_TIOGA_VARIANTS} %cce@=17.0.1 ${PROJECT_TIOGA_DEPS}"
+  extends: .job_on_tioga
+
+# TODO: update cce when vetted
 master:
   variables:
-    SPEC: " +tests %cce@18.0.0"
+    SPEC: " +tests %cce@17.0.1"
   extends: .job_on_tioga

.uberenv_config.json

@@ -4,7 +4,7 @@
     "package_source_dir" : "../..",
     "package_final_phase" : "initconfig",
     "spack_url": "https://github.com/spack/spack.git",
-    "spack_branch": "develop-2024-05-26",
+    "spack_branch": "develop-2024-10-06",
     "spack_activate" : {},
     "spack_configs_path": "scripts/radiuss-spack-configs",
     "spack_packages_path": "scripts/radiuss-spack-configs/packages",

scripts/gitlab/build-and-test.sh

@@ -28,6 +28,7 @@ job_unique_id=${CI_JOB_ID:-""}
 use_dev_shm=${USE_DEV_SHM:-true}
 spack_debug=${SPACK_DEBUG:-false}
 debug_mode=${DEBUG_MODE:-false}
+push_to_registry=${PUSH_TO_REGISTRY:-true}
 
 # REGISTRY_TOKEN allows you to provide your own personal access token to the CI
 # registry. Be sure to set the token with at least read access to the registry.
@@ -53,6 +54,7 @@ then
     echo "~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"
     use_dev_shm=false
     spack_debug=true
+    push_to_registry=false
 fi
 
 if [[ -n ${module_list} ]]
@@ -134,7 +136,7 @@ then
     timed_message "Spack build of dependencies"
     ${uberenv_cmd} --skip-setup-and-env --spec="${spec}" ${prefix_opt}
 
-    if [[ -n ${ci_registry_token} && ${debug_mode} == false ]]
+    if [[ -n ${ci_registry_token} && ${push_to_registry} == true ]]
     then
         timed_message "Push dependencies to buildcache"
         ${spack_cmd} -D ${spack_env_path} buildcache push --only dependencies gitlab_ci
@@ -203,7 +205,7 @@ then
     rm -rf ${build_dir} 2>/dev/null
     mkdir -p ${build_dir} && cd ${build_dir}
 
-    timed_message "Building RAJA"
+    timed_message "Building Caliper"
     if [[ "${truehostname}" == "corona" || "${truehostname}" == "tioga" ]]
     then
         module unload rocm
@@ -245,7 +247,7 @@ then
 
     cd ${build_dir}
 
-    timed_message "Testing RAJA"
+    timed_message "Testing Caliper"
     ctest --output-on-failure -T test 2>&1 | tee tests_output.txt
 
     no_test_str="No tests were found!!!"