Add macOS ML CI stacks (#36586)

* Add macOS ML CI stacks

* torchmeta is no longer maintained and requires ancient PyTorch

* Add MXNet

* update darwin aarch64 stacks

* add darwin-aarch64 scoped config.yaml

* remove unnecessary cleanup job

* fix specifications

* fix labels

* fix labels

* fix indent on tags specification

* no tags for trigger jobs

* try overriding tags in stack spack.yaml

* do not use CI_STACK_CONFIG_SCOPES

* incorporate config:install_tree:root: overrides and compiler defs

* copy relevant ci-scoped config settings directly into stack spack.yaml

* remove build-job-remove

* spack ci generate: add debug flag

* include cdash config directly in stack spack.yaml

* customize build-job script section to avoid absolute paths

* add any-job specification

* tags: use aarch64-macos instead of aarch64

* generate tags: use aarch64-macos instead of aarch64

* do not add morepadding

* use shared mirror; comment out known failures

* remove any-job

* nproc || true

* comment out specs failing due to bazel from cache codesign issue

---------

Co-authored-by: eugeneswalker <eugenesunsetwalker@gmail.com>

share/spack/gitlab/cloud_pipelines/.gitlab-ci.yml

@@ -128,6 +128,38 @@ default:
   extends: [ ".base-job", ".generate-base" ]
   tags: ["spack", "public", "medium", "x86_64"]
 
+.darwin-generate-base:
+  stage: generate
+  script:
+  - export SPACK_DISABLE_LOCAL_CONFIG=1
+  - export SPACK_USER_CACHE_PATH=$(pwd)/_user_cache
+  - uname -a || true
+  - grep -E 'vendor|model name' /proc/cpuinfo 2>/dev/null | sort -u || head -n10 /proc/cpuinfo 2>/dev/null || true
+  - nproc || true
+  - . "./share/spack/setup-env.sh"
+  - spack --version
+  - cd share/spack/gitlab/cloud_pipelines/stacks/${SPACK_CI_STACK_NAME}
+  - spack env activate --without-view .
+  - spack -d ci generate --check-index-only
+      --buildcache-destination "${SPACK_BUILDCACHE_DESTINATION}"
+      --artifacts-root "${CI_PROJECT_DIR}/jobs_scratch_dir"
+      --output-file "${CI_PROJECT_DIR}/jobs_scratch_dir/cloud-ci-pipeline.yml"
+  after_script:
+  - cat /proc/loadavg || true
+  artifacts:
+    paths:
+    - "${CI_PROJECT_DIR}/jobs_scratch_dir"
+  interruptible: true
+  timeout: 60 minutes
+  retry:
+    max: 2
+    when:
+    - always
+
+.darwin-generate:
+  extends: [ ".base-job", ".darwin-generate-base" ]
+
+
 .generate-deprecated:
   extends: [ ".base-job" ]
   stage: generate
@@ -236,124 +268,6 @@ protected-publish:
 #     - artifacts: True
 #       job: my-super-cool-stack-generate
 
-########################################
-#          E4S Mac Stack
-#
-# With no near-future plans to have
-# protected aws runners running mac
-# builds, it seems best to decouple
-# them from the rest of the stacks for
-# the time being.  This way they can
-# still run on UO runners and be signed
-# using the previous approach.
-########################################
-# .e4s-mac:
-#   variables:
-#     SPACK_CI_STACK_NAME: e4s-mac
-#   allow_failure: True
-
-# .mac-pr:
-#   only:
-#   - /^pr[\d]+_.*$/
-#   - /^github\/pr[\d]+_.*$/
-#   variables:
-#     SPACK_BUILDCACHE_DESTINATION: "s3://spack-binaries-prs/${CI_COMMIT_REF_NAME}"
-#     SPACK_PRUNE_UNTOUCHED: "True"
-
-# .mac-protected:
-#   only:
-#   - /^develop$/
-#   - /^releases\/v.*/
-#   - /^v.*/
-#   - /^github\/develop$/
-#   variables:
-#     SPACK_BUILDCACHE_DESTINATION: "s3://spack-binaries/${CI_COMMIT_REF_NAME}/${SPACK_CI_STACK_NAME}"
-
-# .mac-pr-build:
-#   extends: [ ".mac-pr", ".build" ]
-#   variables:
-#     AWS_ACCESS_KEY_ID: ${PR_MIRRORS_AWS_ACCESS_KEY_ID}
-#     AWS_SECRET_ACCESS_KEY: ${PR_MIRRORS_AWS_SECRET_ACCESS_KEY}
-# .mac-protected-build:
-#   extends: [ ".mac-protected", ".build" ]
-#   variables:
-#     AWS_ACCESS_KEY_ID: ${PROTECTED_MIRRORS_AWS_ACCESS_KEY_ID}
-#     AWS_SECRET_ACCESS_KEY: ${PROTECTED_MIRRORS_AWS_SECRET_ACCESS_KEY}
-#     SPACK_SIGNING_KEY: ${PACKAGE_SIGNING_KEY}
-
-# e4s-mac-pr-generate:
-#   extends: [".e4s-mac", ".mac-pr"]
-#   stage: generate
-#   script:
-#     - tmp="$(mktemp -d)"; export SPACK_USER_CONFIG_PATH="$tmp"; export SPACK_USER_CACHE_PATH="$tmp"
-#     - . "./share/spack/setup-env.sh"
-#     - spack --version
-#     - cd share/spack/gitlab/cloud_pipelines/stacks/${SPACK_CI_STACK_NAME}
-#     - spack env activate --without-view .
-#     - spack ci generate --check-index-only
-#       --buildcache-destination "${SPACK_BUILDCACHE_DESTINATION}"
-#       --artifacts-root "${CI_PROJECT_DIR}/jobs_scratch_dir"
-#       --output-file "${CI_PROJECT_DIR}/jobs_scratch_dir/cloud-ci-pipeline.yml"
-#   artifacts:
-#     paths:
-#       - "${CI_PROJECT_DIR}/jobs_scratch_dir"
-#   tags:
-#   - lambda
-#   interruptible: true
-#   retry:
-#     max: 2
-#     when:
-#       - runner_system_failure
-#       - stuck_or_timeout_failure
-#   timeout: 60 minutes
-
-# e4s-mac-protected-generate:
-#   extends: [".e4s-mac", ".mac-protected"]
-#   stage: generate
-#   script:
-#     - tmp="$(mktemp -d)"; export SPACK_USER_CONFIG_PATH="$tmp"; export SPACK_USER_CACHE_PATH="$tmp"
-#     - . "./share/spack/setup-env.sh"
-#     - spack --version
-#     - cd share/spack/gitlab/cloud_pipelines/stacks/${SPACK_CI_STACK_NAME}
-#     - spack env activate --without-view .
-#     - spack ci generate --check-index-only
-#       --artifacts-root "${CI_PROJECT_DIR}/jobs_scratch_dir"
-#       --output-file "${CI_PROJECT_DIR}/jobs_scratch_dir/cloud-ci-pipeline.yml"
-#   artifacts:
-#     paths:
-#       - "${CI_PROJECT_DIR}/jobs_scratch_dir"
-#   tags:
-#   - omicron
-#   interruptible: true
-#   retry:
-#     max: 2
-#     when:
-#       - runner_system_failure
-#       - stuck_or_timeout_failure
-#   timeout: 60 minutes
-
-# e4s-mac-pr-build:
-#   extends: [ ".e4s-mac", ".mac-pr-build" ]
-#   trigger:
-#     include:
-#       - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
-#         job: e4s-mac-pr-generate
-#     strategy: depend
-#   needs:
-#     - artifacts: True
-#       job: e4s-mac-pr-generate
-
-# e4s-mac-protected-build:
-#   extends: [ ".e4s-mac", ".mac-protected-build" ]
-#   trigger:
-#     include:
-#       - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
-#         job: e4s-mac-protected-generate
-#     strategy: depend
-#   needs:
-#     - artifacts: True
-#       job: e4s-mac-protected-generate
-
 ########################################
 # E4S pipeline
 ########################################
@@ -762,6 +676,28 @@ ml-linux-x86_64-rocm-build:
     - artifacts: True
       job: ml-linux-x86_64-rocm-generate
 
+########################################
+# Machine Learning - Darwin aarch64 (MPS)
+########################################
+.ml-darwin-aarch64-mps:
+  variables:
+    SPACK_CI_STACK_NAME: ml-darwin-aarch64-mps
+
+ml-darwin-aarch64-mps-generate:
+  tags: [ "macos-ventura", "apple-clang-14", "aarch64-macos" ]
+  extends: [ ".ml-darwin-aarch64-mps", ".darwin-generate"]
+
+ml-darwin-aarch64-mps-build:
+  extends: [ ".ml-darwin-aarch64-mps", ".build" ]
+  trigger:
+    include:
+      - artifact: jobs_scratch_dir/cloud-ci-pipeline.yml
+        job: ml-darwin-aarch64-mps-generate
+    strategy: depend
+  needs:
+    - artifacts: True
+      job: ml-darwin-aarch64-mps-generate
+
 ########################################
 # Deprecated CI testing
 ########################################

share/spack/gitlab/cloud_pipelines/configs/darwin/aarch64/compilers.yaml

@@ -0,0 +1,27 @@
+compilers:
+- compiler:
+    spec: apple-clang@14.0.0
+    paths:
+      cc: /usr/bin/clang
+      cxx: /usr/bin/clang++
+      f77: /opt/homebrew/bin/gfortran
+      fc: /opt/homebrew/bin/gfortran
+    flags: {}
+    operating_system: ventura
+    target: aarch64
+    modules: []
+    environment: {}
+    extra_rpaths: []
+- compiler:
+    spec: gcc@12.2.0
+    paths:
+      cc: /opt/homebrew/bin/gcc-12
+      cxx: /opt/homebrew/bin/g++-12
+      f77: /opt/homebrew/bin/gfortran-12
+      fc: /opt/homebrew/bin/gfortran-12
+    flags: {}
+    operating_system: ventura
+    target: aarch64
+    modules: []
+    environment: {}
+    extra_rpaths: []

share/spack/gitlab/cloud_pipelines/configs/darwin/aarch64/config.yaml

@@ -0,0 +1,3 @@
+config:
+  install_tree:
+    root: $spack/opt/spack

share/spack/gitlab/cloud_pipelines/configs/darwin/ci.yaml

@@ -1,14 +0,0 @@
-ci:
-  pipeline-gen:
-  - build-job:
-      script: |
-        - tmp="$(mktemp -d)"; export SPACK_USER_CONFIG_PATH="$tmp"; export SPACK_USER_CACHE_PATH="$tmp"
-        - . "./share/spack/setup-env.sh"
-        - spack --version
-        - spack arch
-        - cd ${SPACK_CONCRETE_ENV_DIR}
-        - spack env activate --without-view .
-        - spack config add "config:install_tree:projections:${SPACK_JOB_SPEC_PKG_NAME}:'morepadding/{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}'"
-        - mkdir -p ${SPACK_ARTIFACTS_ROOT}/user_data
-        - spack --color=always --backtrace ci rebuild > >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_out.txt) 2> >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_err.txt >&2)
-      tags: ["lambda"]

share/spack/gitlab/cloud_pipelines/stacks/e4s-mac/spack.yaml

@@ -1,40 +0,0 @@
-spack:
-  view: false
-  packages:
-    all:
-      compiler: [apple-clang@13.1.6]
-      target: [m1]
-
-  definitions:
-  - easy_specs:
-    - berkeley-db
-    - ncurses
-    - gcc
-    - py-jupyterlab
-    - py-scipy
-    - py-matplotlib
-    - py-pandas
-
-  - arch:
-    - '%apple-clang@13.1.6 target=m1'
-
-  specs:
-
-  - matrix:
-    - - $easy_specs
-    - - $arch
-
-  mirrors: { "mirror": "s3://spack-binaries/develop/e4s-mac" }
-
-  ci:
-    pipeline-gen:
-    - cleanup-job:
-        before_script: |
-          - export SPACK_USER_CACHE_PATH=$(pwd)/.spack-user-cache
-          - export SPACK_USER_CONFIG_PATH=$(pwd)/.spack-user-config
-          - . "./share/spack/setup-env.sh"
-          - spack --version
-        tags: [lambda]
-
-  cdash:
-    build-group: E4S Mac

share/spack/gitlab/cloud_pipelines/stacks/ml-darwin-aarch64-mps/spack.yaml

@@ -0,0 +1,139 @@
+spack:
+  view: false
+
+  concretizer:
+    unify: false
+    reuse: false
+
+  config:
+    concretizer: clingo
+    db_lock_timeout: 120
+    install_tree:
+      root: $spack/opt/spack
+      padded_length: 256
+      projections:
+        all: '{architecture}/{compiler.name}-{compiler.version}/{name}-{version}-{hash}'
+
+  packages:
+    all:
+      require: target=aarch64
+      variants: +mps~cuda~rocm
+    mpi:
+      require: openmpi
+
+  specs:
+  # Hugging Face
+  - py-transformers
+
+  # JAX
+  - py-jax
+  # - py-jaxlib     # bazel codesign
+
+  # Keras
+  - py-keras-applications
+  - py-keras-preprocessing
+  - py-keras2onnx
+  # - py-keras      # bazel codesign
+
+  # MXNet
+  - mxnet
+
+  # PyTorch
+  - py-botorch
+  - py-gpytorch
+  - py-pytorch-gradual-warmup-lr
+  - py-segmentation-models-pytorch
+  - py-timm
+  - py-torch
+  - py-torch-cluster
+  - py-torch-geometric
+  - py-torch-sparse
+  - py-torchdata
+  - py-torchfile
+  - py-torchgeo
+  - py-torchvision
+
+  # scikit-learn
+  - py-scikit-learn
+  - py-scikit-learn-extra
+
+  # TensorBoard
+  - py-tensorboard
+  - py-tensorboard-data-server
+  - py-tensorboard-plugin-wit
+  - py-tensorboardx
+
+  # TensorFlow
+  # - py-tensorflow               # bazel codesign
+  # - py-tensorflow-datasets      # bazel codesign
+  # - py-tensorflow-hub           # bazel codesign
+  # - py-tensorflow-metadata      # bazel codesign
+  # - py-tensorflow-estimator     # bazel codesign
+  # - py-tensorflow-probability   # py-dm-tree due to bazel codesign
+
+  # XGBoost
+  - py-xgboost
+  - xgboost
+
+  # ERRORS
+  # - py-efficientnet-pytorch     # py-torch
+  # - py-horovod                  # py-torch
+  # - py-kornia                   # py-torch
+  # - py-lightning                # py-torch
+  # - py-pytorch-lightning        # py-torch
+  # - py-torch-nvidia-apex        # py-torch
+  # - py-torch-scatter            # py-torch
+  # - py-torch-spline-conv        # py-torch
+  # - py-torchaudio               # py-torchaudio
+  # - py-torchmetrics             # py-torch
+  # - py-torchtext                # py-torchtext
+  # - py-vector-quantize-pytorch  # py-torch
+  # - r-xgboost                   # r
+
+  mirrors: { "mirror": "s3://spack-binaries/develop/ml-darwin-aarch64-cpu" }
+
+  ci:
+    pipeline-gen:
+    - build-job-remove:
+        image: no-image
+        tags: [spack, public]
+    - build-job:
+        tags: [ "macos-ventura", "apple-clang-14", "aarch64-macos" ]
+        script::
+        - - spack compiler find
+          - cd ${SPACK_CONCRETE_ENV_DIR}
+          - spack env activate --without-view .
+          - if [ -n "$SPACK_BUILD_JOBS" ]; then spack config add "config:build_jobs:$SPACK_BUILD_JOBS"; fi
+          - mkdir -p ${SPACK_ARTIFACTS_ROOT}/user_data
+          # AWS runners mount E4S public key (verification), UO runners mount public/private (signing/verification)
+          - if [[ -r /mnt/key/e4s.gpg ]]; then spack gpg trust /mnt/key/e4s.gpg; fi
+          # UO runners mount intermediate ci public key (verification), AWS runners mount public/private (signing/verification)
+          - if [[ -r /mnt/key/intermediate_ci_signing_key.gpg ]]; then spack gpg trust /mnt/key/intermediate_ci_signing_key.gpg; fi
+          - if [[ -r /mnt/key/spack_public_key.gpg ]]; then spack gpg trust /mnt/key/spack_public_key.gpg; fi
+          - spack --color=always --backtrace ci rebuild --tests > >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_out.txt) 2> >(tee ${SPACK_ARTIFACTS_ROOT}/user_data/pipeline_err.txt >&2)
+        after_script:
+        - - cat /proc/loadavg || true
+    - signing-job:
+        image: { "name": "ghcr.io/spack/notary:latest", "entrypoint": [""] }
+        tags: ["aws"]
+        script:
+        - - aws s3 sync --exclude "*" --include "*spec.json*" ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache /tmp
+          - /sign.sh
+          - aws s3 sync --exclude "*" --include "*spec.json.sig*" /tmp ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache
+          - aws s3 cp /tmp/public_keys ${SPACK_REMOTE_MIRROR_OVERRIDE}/build_cache/_pgp --recursive --exclude "*" --include "*.pub"
+    - any-job:
+        image: "ghcr.io/spack/e4s-ubuntu-18.04:v2021-10-18"
+        tags: ["spack"]
+        before_script:
+        - - uname -a || true
+          - grep -E "vendor|model name" /proc/cpuinfo 2>/dev/null | sort -u || head -n10 /proc/cpuinfo 2>/dev/null || true
+          - nproc || true
+        - - . "./share/spack/setup-env.sh"
+          - spack --version
+          - spack arch
+
+  cdash:
+    build-group: Machine Learning MPS
+    url: https://cdash.spack.io
+    project: Spack Testing
+    site: Cloud Gitlab Infrastructure