Move "platforms" CI jobs into their own workflow files. (iree-org#17979)

This moves workflow jobs that build the full project (including the compiler) from `ci.yml` to their own workflow files: - `build_test_all_arm64` --> `.github/workflows/ci_linux_arm64_clang.yml` - `build_test_all_windows` --> `.github/workflows/ci_windows_x86_msvc.yml` - `build_test_all_macos_arm64` --> none - `build_test_all_macos_x86_64` --> `.github/workflows/ci_macos_x86_clang.yml` With one workflow file per platform, these are the tradeoffs compared to a single workflow file: Pros: + Easy to tune which triggers affect each workflow * See https://github.com/Tencent/ncnn/blob/master/.github/workflows/windows-clang.yml for some inspiration. Each workflow can independently define `paths:` to trigger on. We have support for that via [`build_tools/github_actions/configure_ci.py`](https://github.com/iree-org/iree/blob/main/build_tools/github_actions/configure_ci.py), but this also allows for adjusting schedule triggers too. + Easy to fork a workflow for a new platform, less merge conflicts in a single large file + Easy to cancel a single job without affecting others from the same file + Easy to disable/enable a workflow in isolation from the UI if there are runner issues + Low impact radius if/when a single platform has runner issues Cons: - If running on PRs, checks are spread out a lot - Hard to mark multiple/all jobs as required checks (if we care) - More code duplication between files (reusable workflows can help here... maybe, could also lean on CMake presets and build scripts) Practically at this point in time, these workflows _were_ running on all commits pushed to the `main` branch but they were disabled for being too slow or requiring runners that went offline. Now the workflows will run on a nightly schedule and they can be disabled individually from the GitHub UI as needed. TBD how we'll make these workflows available for opt-in on presubmit. We could add the `pull_request` trigger back, trigger if some labels are added, etc. skip-ci: adding new workflows, existing ones are unaffected
nod-ai · Jul 22, 2024 · 7463cb4 · 7463cb4
1 parent 108e050
commit 7463cb4
Show file tree

Hide file tree

Showing 7 changed files with 174 additions and 248 deletions.
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -65,185 +65,6 @@ jobs:
       write-caches: ${{ needs.setup.outputs.write-caches }}
       run-tests: true
 
-  # Disabled since we don't have any runners for arm64.
-  # As of July 2024, GitHub has "large" arm64 runners available in beta at cost.
-  # If these get included in the free/standard tier or if a project member
-  # sponsors their use then we could re-enable this job.
-  # build_test_all_arm64:
-  #   needs: setup
-  #   if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'build_test_all_arm64')
-  #   runs-on:
-  #     - self-hosted # must come first
-  #     - runner-group=${{ needs.setup.outputs.runner-group }}
-  #     - environment=${{ needs.setup.outputs.runner-env }}
-  #     - arm64
-  #     - os-family=Linux
-  #   env:
-  #     BUILD_DIR: build-arm64
-  #   steps:
-  #     - name: "Checking out repository"
-  #       uses: actions/[email protected]
-  #       with:
-  #         submodules: true
-  #     - name: "Building IREE"
-  #       env:
-  #         IREE_WRITE_REMOTE_CCACHE: ${{ needs.setup.outputs.write-caches }}
-  #       run: |
-  #         ./build_tools/github_actions/docker_run.sh \
-  #           --env "IREE_CCACHE_GCP_TOKEN=$(gcloud auth application-default print-access-token)" \
-  #           --env "IREE_WRITE_REMOTE_CCACHE=${IREE_WRITE_REMOTE_CCACHE}" \
-  #           --env "CCACHE_NAMESPACE=gcr.io/iree-oss/base-arm64@sha256:9daa1cdbbf12da8527319ece76a64d06219e04ecb99a4cff6e6364235ddf6c59" \
-  #           --env "IREE_BUILD_SETUP_PYTHON_VENV=${BUILD_DIR}/.venv" \
-  #           gcr.io/iree-oss/base-arm64@sha256:9daa1cdbbf12da8527319ece76a64d06219e04ecb99a4cff6e6364235ddf6c59 \
-  #           ./build_tools/cmake/build_all.sh \
-  #           "${BUILD_DIR}"
-  #     - name: "Testing IREE"
-  #       run: |
-  #         ./build_tools/github_actions/docker_run.sh \
-  #           --env "IREE_ARM_SME_QEMU_AARCH64_BIN=/usr/bin/qemu-aarch64" \
-  #           gcr.io/iree-oss/base-arm64@sha256:9daa1cdbbf12da8527319ece76a64d06219e04ecb99a4cff6e6364235ddf6c59 \
-  #           ./build_tools/cmake/ctest_all.sh \
-  #           "${BUILD_DIR}"
-
-  # Disabled since
-  #   * windows-2022 is too slow
-  #   * windows-2022-64core is too expensive (and takes ~8 minutes just to checkout the repo)
-  #   * we don't have self-hosted Windows runners with enough cores yet
-  # build_test_all_windows:
-  #   needs: setup
-  #   if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'build_test_all_windows')
-  #   # TODO(saienduri): switch to self-hosted
-  #   runs-on: windows-2022
-  #   defaults:
-  #     run:
-  #       shell: bash
-  #   env:
-  #     BUILD_DIR: build-windows
-  #   steps:
-  #     - name: "Checking out repository"
-  #       uses: actions/[email protected]
-  #       with:
-  #         submodules: true
-  #     - id: "gcp-auth"
-  #       name: "Authenticating to Google Cloud"
-  #       if: needs.setup.outputs.write-caches == 1
-  #       uses: "google-github-actions/auth@v1"
-  #       with:
-  #         token_format: "access_token"
-  #         credentials_json: "${{ secrets.IREE_OSS_GITHUB_RUNNER_BASIC_TRUST_SERVICE_ACCOUNT_KEY }}"
-  #         create_credentials_file: false
-  #     - name: "Setting up Python"
-  #       uses: actions/[email protected]
-  #       with:
-  #         python-version: "3.10" # Needs pybind >= 2.10.1 for Python >= 3.11
-  #     - name: "Installing Python packages"
-  #       run: |
-  #         python3 -m venv .venv
-  #         .venv/Scripts/activate.bat
-  #         python3 -m pip install -r runtime/bindings/python/iree/runtime/build_requirements.txt
-  #     - name: "Installing requirements"
-  #       run: choco install ccache --yes
-  #     - name: "Configuring MSVC"
-  #       uses: ilammy/[email protected]
-  #     # Finally: build and run tests.
-  #     - name: "Building IREE"
-  #       env:
-  #         IREE_WRITE_REMOTE_CCACHE: ${{ needs.setup.outputs.write-caches }}
-  #         IREE_CCACHE_GCP_TOKEN: ${{ steps.gcp-auth.outputs.access_token }}
-  #         CCACHE_NAMESPACE: github-windows-2022
-  #       run: ./build_tools/cmake/build_all.sh "${BUILD_DIR}"
-  #     - name: "Testing IREE"
-  #       run: ./build_tools/cmake/ctest_all.sh "${BUILD_DIR}"
-
-  # Disabled while runner is offline.
-  # build_test_all_macos_arm64:
-  #   needs: setup
-  #   if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'build_test_all_macos_arm64')
-  #   runs-on:
-  #     - ${{ github.repository == 'iree-org/iree' && 'self-hosted' || 'macos-12' }} # must come first
-  #     - runner-group=postsubmit
-  #     - os-family=macOS
-  #   env:
-  #     BUILD_DIR: build-macos
-  #   steps:
-  #     - name: "Checking out repository"
-  #       uses: actions/[email protected]
-  #     - name: "Updating git submodules"
-  #       run: git submodule update --init --jobs 8 --depth 1
-  #     - name: "Installing Python packages"
-  #       run: |
-  #         python3 -m venv .venv
-  #         source .venv/bin/activate
-  #         python3 -m pip install -r runtime/bindings/python/iree/runtime/build_requirements.txt
-  #     - name: "Building IREE"
-  #       env:
-  #         IREE_READ_REMOTE_CCACHE: 0
-  #         IREE_WRITE_REMOTE_CCACHE: 0
-  #         IREE_READ_LOCAL_CCACHE: 1
-  #         IREE_WRITE_LOCAL_CCACHE: ${{ needs.setup.outputs.write-caches }}
-  #         # We'll remove the GITHUB_WORKSPACE directory after the job.
-  #         # Persist the cache by storing in the parent directory.
-  #         CCACHE_DIR: ${{ github.workspace }}/../.ccache
-  #         CCACHE_MAXSIZE: 30G
-  #       run: |
-  #         source .venv/bin/activate
-  #         bash ./build_tools/cmake/build_all.sh "${BUILD_DIR}"
-  #     - name: "Testing IREE"
-  #       run: |
-  #         source .venv/bin/activate
-  #         IREE_METAL_DISABLE=0 bash ./build_tools/cmake/ctest_all.sh "${BUILD_DIR}"
-
-  build_test_all_macos_x86_64:
-    needs: setup
-    if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'build_test_all_macos_x86_64')
-    runs-on: macos-13
-    env:
-      BUILD_DIR: build-macos
-    defaults:
-      run:
-        shell: bash
-    steps:
-      - name: "Checking out repository"
-        uses: actions/[email protected]
-      - id: "gcp-auth"
-        name: "Authenticating to Google Cloud"
-        if: needs.setup.outputs.write-caches == 1
-        uses: "google-github-actions/auth@v1"
-        with:
-          token_format: "access_token"
-          credentials_json: "${{ secrets.IREE_OSS_GITHUB_RUNNER_BASIC_TRUST_SERVICE_ACCOUNT_KEY }}"
-          create_credentials_file: false
-      - name: "Updating git submodules"
-        run: git submodule update --init --jobs 8 --depth 1
-      # There are multiple versions of Xcode and SDKs installed on the macOS runner.
-      # Select the latest Xcode app to enable using Metal offline toolchain.
-      - name: "Update Xcode command line tools path"
-        run: |
-          sudo xcode-select --switch /Applications/Xcode.app/Contents/Developer
-          xcrun metal --version
-          xcrun metallib --version
-      - name: "Setting up Python"
-        uses: actions/[email protected]
-        with:
-          python-version: "3.10"
-          cache: "pip"
-      - name: "Installing Python packages"
-        run: pip install -r runtime/bindings/python/iree/runtime/build_requirements.txt
-      - name: "Installing requirements"
-        # We need coreutils for `realpath` used in scripts.
-        # We need bash because the default one on macOS is fairly old and lacks
-        # features we use in scripts.
-        run: brew install ninja ccache coreutils bash
-      # Finally: build and run tests.
-      - name: "Building IREE"
-        env:
-          IREE_WRITE_REMOTE_CCACHE: ${{ needs.setup.outputs.write-caches }}
-          IREE_CCACHE_GCP_TOKEN: ${{ steps.gcp-auth.outputs.access_token }}
-          CCACHE_NAMESPACE: github-macos-13
-        run: bash ./build_tools/cmake/build_all.sh "${BUILD_DIR}"
-      - name: "Testing IREE"
-        run: bash ./build_tools/cmake/ctest_all.sh "${BUILD_DIR}"
-
   build_test_all_bazel:
     needs: setup
     if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'build_test_all_bazel')
@@ -984,12 +805,6 @@ jobs:
       - build_all
       - build_test_all_bazel
 
-      # Platforms
-      # - build_test_all_arm64
-      # - build_test_all_windows
-      # - build_test_all_macos_arm64
-      - build_test_all_macos_x86_64
-
       # Accelerators
       - test_nvidia_gpu
       # - test_nvidia_a100

diff --git a/.github/workflows/ci_linux_arm64_clang.yml b/.github/workflows/ci_linux_arm64_clang.yml
@@ -0,0 +1,63 @@
+# Copyright 2024 The IREE Authors
+#
+# Licensed under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+name: CI - Linux arm64 clang
+
+on:
+  schedule:
+    # Weekday mornings at 09:15 UTC = 01:15 PST (UTC - 8).
+    - cron: "15 9 * * 1-5"
+  workflow_dispatch:
+
+concurrency:
+  # A PR number if a pull request and otherwise the commit hash. This cancels
+  # queued and in-progress runs for the same PR (presubmit) or commit
+  # (postsubmit). The workflow name is prepended to avoid conflicts between
+  # different workflows.
+  group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
+  cancel-in-progress: true
+
+jobs:
+  setup:
+    uses: ./.github/workflows/setup.yml
+
+  # TODO: Switch to GitHub-hosted arm64 runners when available?
+  # TODO: Switch runs-on labels to use different (non-GCP) self-hosted runners?
+  linux_arm64_clang:
+    needs: setup
+    if: contains(fromJson(needs.setup.outputs.enabled-jobs), 'build_test_all_arm64')
+    runs-on:
+      - self-hosted # must come first
+      - runner-group=${{ needs.setup.outputs.runner-group }}
+      - environment=${{ needs.setup.outputs.runner-env }}
+      - arm64
+      - os-family=Linux
+    env:
+      BUILD_DIR: build-arm64
+    steps:
+      - name: "Checking out repository"
+        uses: actions/[email protected]
+        with:
+          submodules: true
+      - name: "Building IREE"
+        env:
+          IREE_WRITE_REMOTE_CCACHE: ${{ needs.setup.outputs.write-caches }}
+        run: |
+          ./build_tools/github_actions/docker_run.sh \
+            --env "IREE_CCACHE_GCP_TOKEN=$(gcloud auth application-default print-access-token)" \
+            --env "IREE_WRITE_REMOTE_CCACHE=${IREE_WRITE_REMOTE_CCACHE}" \
+            --env "CCACHE_NAMESPACE=gcr.io/iree-oss/base-arm64@sha256:9daa1cdbbf12da8527319ece76a64d06219e04ecb99a4cff6e6364235ddf6c59" \
+            --env "IREE_BUILD_SETUP_PYTHON_VENV=${BUILD_DIR}/.venv" \
+            gcr.io/iree-oss/base-arm64@sha256:9daa1cdbbf12da8527319ece76a64d06219e04ecb99a4cff6e6364235ddf6c59 \
+            ./build_tools/cmake/build_all.sh \
+            "${BUILD_DIR}"
+      - name: "Testing IREE"
+        run: |
+          ./build_tools/github_actions/docker_run.sh \
+            --env "IREE_ARM_SME_QEMU_AARCH64_BIN=/usr/bin/qemu-aarch64" \
+            gcr.io/iree-oss/base-arm64@sha256:9daa1cdbbf12da8527319ece76a64d06219e04ecb99a4cff6e6364235ddf6c59 \
+            ./build_tools/cmake/ctest_all.sh \
+            "${BUILD_DIR}"
diff --git a/.github/workflows/ci_macos_x86_clang.yml b/.github/workflows/ci_macos_x86_clang.yml
@@ -0,0 +1,58 @@
+# Copyright 2024 The IREE Authors
+#
+# Licensed under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+name: CI - macOS x86 clang
+
+on:
+  schedule:
+    # Weekday mornings at 09:15 UTC = 01:15 PST (UTC - 8).
+    - cron: "15 9 * * 1-5"
+  workflow_dispatch:
+
+concurrency:
+  # A PR number if a pull request and otherwise the commit hash. This cancels
+  # queued and in-progress runs for the same PR (presubmit) or commit
+  # (postsubmit). The workflow name is prepended to avoid conflicts between
+  # different workflows.
+  group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
+  cancel-in-progress: true
+
+jobs:
+  macos_x86_clang:
+    runs-on: macos-13
+    env:
+      BUILD_DIR: build-macos
+    defaults:
+      run:
+        shell: bash
+    steps:
+      - name: "Checking out repository"
+        uses: actions/[email protected]
+      - name: "Updating git submodules"
+        run: git submodule update --init --jobs 8 --depth 1
+      # There may be multiple versions of Xcode and SDKs installed.
+      # Select the latest Xcode app to enable using Metal offline toolchain.
+      - name: "Update Xcode command line tools path"
+        run: |
+          sudo xcode-select --switch /Applications/Xcode.app/Contents/Developer
+          xcrun metal --version
+          xcrun metallib --version
+      - name: "Setting up Python"
+        uses: actions/[email protected]
+        with:
+          python-version: "3.10"
+          cache: "pip"
+      - name: "Installing Python packages"
+        run: pip install -r runtime/bindings/python/iree/runtime/build_requirements.txt
+      - name: "Installing requirements"
+        # We need coreutils for `realpath` used in scripts.
+        # We need bash because the default one on macOS is fairly old and lacks
+        # features we use in scripts.
+        run: brew install ninja ccache coreutils bash
+      - name: "Building IREE"
+        run: bash ./build_tools/cmake/build_all.sh "${BUILD_DIR}"
+      - name: "Testing IREE"
+        run: bash ./build_tools/cmake/ctest_all.sh "${BUILD_DIR}"
diff --git a/.github/workflows/ci_windows_x86_msvc.yml b/.github/workflows/ci_windows_x86_msvc.yml
@@ -0,0 +1,52 @@
+# Copyright 2024 The IREE Authors
+#
+# Licensed under the Apache License v2.0 with LLVM Exceptions.
+# See https://llvm.org/LICENSE.txt for license information.
+# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+
+name: CI - Windows x86 MSVC
+
+on:
+  schedule:
+    # Weekday mornings at 09:15 UTC = 01:15 PST (UTC - 8).
+    - cron: "15 9 * * 1-5"
+  workflow_dispatch:
+
+concurrency:
+  # A PR number if a pull request and otherwise the commit hash. This cancels
+  # queued and in-progress runs for the same PR (presubmit) or commit
+  # (postsubmit). The workflow name is prepended to avoid conflicts between
+  # different workflows.
+  group: ${{ github.workflow }}-${{ github.event.number || github.sha }}
+  cancel-in-progress: true
+
+jobs:
+  windows_x86_msvc:
+    runs-on: windows-2022
+    defaults:
+      run:
+        shell: bash
+    env:
+      BUILD_DIR: build-windows
+    steps:
+      - name: "Checking out repository"
+        uses: actions/[email protected]
+        with:
+          submodules: true
+      - name: "Setting up Python"
+        uses: actions/[email protected]
+        with:
+          python-version: "3.10" # Needs pybind >= 2.10.1 for Python >= 3.11
+      - name: "Installing Python packages"
+        run: |
+          python3 -m venv .venv
+          .venv/Scripts/activate.bat
+          python3 -m pip install -r runtime/bindings/python/iree/runtime/build_requirements.txt
+      - name: "Installing requirements"
+        run: choco install ccache --yes
+      - name: "Configuring MSVC"
+        uses: ilammy/[email protected]
+      - name: "Building IREE"
+        run: ./build_tools/cmake/build_all.sh "${BUILD_DIR}"
+      - name: "Testing IREE"
+        run: ./build_tools/cmake/ctest_all.sh "${BUILD_DIR}"
diff --git a/build_tools/cmake/ctest_all.sh b/build_tools/cmake/ctest_all.sh
@@ -123,6 +123,7 @@ if [[ "${OSTYPE}" =~ ^msys ]]; then
     "iree/tests/e2e/tensor_ops/check_vmvx_ukernel_local-task_unpack.mlir"
     # TODO(#11070): Fix argument/result signature mismatch
     "iree/tests/e2e/tosa_ops/check_vmvx_local-sync_microkernels_fully_connected.mlir"
+    "iree/tests/e2e/tosa_ops/check_vmvx_local-sync_microkernels_matmul.mlir"
   )
 elif [[ "${OSTYPE}" =~ ^darwin ]]; then
   excluded_tests+=(