name cleanup + add H100 runner + install curl

NVIDIA · Dec 16, 2024 · 9a694fa · 9a694fa
1 parent 61813fa
commit 9a694fa
Show file tree

Hide file tree

Showing 3 changed files with 54 additions and 34 deletions.
diff --git a/.github/actions/fetch_ctk/action.yml b/.github/actions/fetch_ctk/action.yml
@@ -19,6 +19,26 @@ runs:
         echo "CTK_CACHE_KEY=mini-ctk-${{ inputs.cuda-version }}-${{ inputs.host-platform }}" >> $GITHUB_ENV
         echo "CTK_CACHE_FILENAME=mini-ctk-${{ inputs.cuda-version }}-${{ inputs.host-platform }}.tar.gz" >> $GITHUB_ENV
 
+    - name: Install dependencies
+      shell: bash --noprofile --norc -xeuo pipefail {0}
+      run: |
+        if (command -v curl 2>&1 >/dev/null) && (command -v zstd 2>&1 >/dev/null); then
+          echo "All dependencies are found. Do nothing."
+          exit 0
+        fi
+        if ! command -v sudo 2>&1 >/dev/null; then
+          if [[ $EUID == 0 ]]; then
+            alias SUDO=""
+          else
+            echo "The following oprations require root access."
+            exit 1
+          fi
+        else
+          alias SUDO="sudo"
+        fi
+        SUDO apt update
+        SUDO apt install -y zstd curl
+
     - name: Download CTK cache
       id: ctk-get-cache
       uses: actions/cache/restore@v4
@@ -32,22 +52,22 @@ runs:
       if: ${{ steps.ctk-get-cache.outputs.cache-hit != 'true' }}
       shell: bash --noprofile --norc -xeuo pipefail {0}
       run: |
-        CUDA_PATH="$(pwd)/cuda_toolkit"
+        CUDA_PATH="./cuda_toolkit"
         mkdir $CUDA_PATH
 
         # The binary archives (redist) are guaranteed to be updated as part of the release posting.
         CTK_BASE_URL="https://developer.download.nvidia.com/compute/cuda/redist/"
         CTK_JSON_URL="$CTK_BASE_URL/redistrib_${{ inputs.cuda-version }}.json"
         if [[ "${{ inputs.host-platform }}" == linux* ]]; then
-          if [[ "${{ inputs.host-platform }}" == "linux-x64" ]]; then
+          if [[ "${{ inputs.host-platform }}" == "linux-64" ]]; then
             CTK_SUBDIR="linux-x86_64"
           elif [[ "${{ inputs.host-platform }}" == "linux-aarch64" ]]; then
             CTK_SUBDIR="linux-sbsa"
           fi
           function extract() {
             tar -xvf $1 -C $CUDA_PATH --strip-components=1
           }
-        elif [[ "${{ inputs.host-platform }}" == "win-x64" ]]; then
+        elif [[ "${{ inputs.host-platform }}" == "win-64" ]]; then
           CTK_SUBDIR="windows-x86_64"
           function extract() {
             _TEMP_DIR_=$(mktemp -d)
@@ -102,12 +122,13 @@ runs:
       shell: bash --noprofile --norc -xeuo pipefail {0}
       run: |
         ls -l
-        CUDA_PATH="$(pwd)/cuda_toolkit"
+        CUDA_PATH="./cuda_toolkit"
         tar -xzvf $CTK_CACHE_FILENAME
         ls -l $CUDA_PATH
         if [ ! -d "$CUDA_PATH/include" ]; then
           exit 1
         fi
+        CUDA_PATH=$(realpath ${CUDA_PATH})
 
         echo "CUDA_PATH=${CUDA_PATH}" >> $GITHUB_ENV
         echo "${CUDA_PATH}/bin" >> $GITHUB_PATH

diff --git a/.github/workflows/ci-gh.yml b/.github/workflows/ci-gh.yml
@@ -1,4 +1,4 @@
-name: "CI"
+name: CI
 
 concurrency:
   group: ${{ startsWith(github.ref_name, 'main') && format('unique-{0}', github.run_id) || format('ci-build-and-test-on-{0}-from-{1}', github.event_name, github.ref_name) }}
@@ -12,7 +12,6 @@ on:
 
 jobs:
   ci:
-    name: "CI"
     uses:
       ./.github/workflows/gh-build-and-test.yml
     secrets: inherit
diff --git a/.github/workflows/gh-build-and-test.yml b/.github/workflows/gh-build-and-test.yml
@@ -5,29 +5,28 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        # TODO: align host-platform names with conda convention
         host-platform:
-          - linux-x64
+          - linux-64
           - linux-aarch64
-          - win-x64
+          - win-64
         python-version:
-          - "3.13"
-          - "3.12"
-          - "3.11"
-          - "3.10"
+#          - "3.13"
+#          - "3.12"
+#          - "3.11"
+#          - "3.10"
           - "3.9"
         cuda-version:
           # Note: this is for build-time only.
           - "12.6.2"
-    name: Build (${{ matrix.host-platform }}, Python "${{ matrix.python-version }}")
+    name: Build (${{ matrix.host-platform }}, Python ${{ matrix.python-version }}, CUDA ${{ matrix.cuda-version }})
     if: ${{ github.repository_owner == 'nvidia' }}
     permissions:
       id-token: write # This is required for configure-aws-credentials
       contents: read  # This is required for actions/checkout
-    runs-on: ${{ (matrix.host-platform == 'linux-x64' && 'linux-amd64-cpu8') ||
+    runs-on: ${{ (matrix.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
                  (matrix.host-platform == 'linux-aarch64' && 'linux-arm64-cpu8') ||
-                 (matrix.host-platform == 'win-x64' && 'windows-2019') }}
-               #  (matrix.host-platform == 'win-x64' && 'windows-amd64-cpu8') }}
+                 (matrix.host-platform == 'win-64' && 'windows-2019') }}
+               #  (matrix.host-platform == 'win-64' && 'windows-amd64-cpu8') }}
     outputs:
       BUILD_CTK_VER: ${{ steps.pass_env.outputs.CUDA_VERSION }}
     steps:
@@ -166,31 +165,39 @@ jobs:
   test:
     strategy:
       fail-fast: false
+      # TODO: add driver version here
       matrix:
-        # TODO: align host-platform names with conda convention
         host-platform:
-          - linux-x64
+          - linux-64
           - linux-aarch64
           # TODO: enable testing once win-64 GPU runners are up
-          # - win-x64
+          # - win-64
         python-version:
-          - "3.13"
-          - "3.12"
-          - "3.11"
-          - "3.10"
+#          - "3.13"
+#          - "3.12"
+#          - "3.11"
+#          - "3.10"
           - "3.9"
         cuda-version:
           # Note: this is for test-time only.
           - "12.6.2"
           - "12.0.1"
           - "11.8.0"
-    name: Test (${{ matrix.host-platform }}, CUDA ${{ matrix.cuda-version }}, Python "${{ matrix.python-version }}")
+        runner:
+          - default
+        include:
+          - host-platform: linux-64
+            python-version: "3.12"
+            cuda-version: "12.6.2"
+            runner: H100
+    name: Test (${{ matrix.host-platform }}, Python ${{ matrix.python-version }}, CUDA ${{ matrix.cuda-version }}, Runner ${{ matrix.runner }})
     if: ${{ (github.repository_owner == 'nvidia') }}
     permissions:
       id-token: write # This is required for configure-aws-credentials
       contents: read  # This is required for actions/checkout
-    runs-on: ${{ (matrix.host-platform == 'linux-x64' && 'linux-amd64-gpu-v100-latest-1') ||
-                 (matrix.host-platform == 'linux-aarch64' && 'linux-arm64-gpu-a100-latest-1') }}
+    runs-on: ${{ (matrix.runner == 'default' && matrix.host-platform == 'linux-64' && 'linux-amd64-gpu-v100-latest-1') ||
+                 (matrix.runner == 'default' && matrix.host-platform == 'linux-aarch64' && 'linux-arm64-gpu-a100-latest-1') ||
+                 (matrix.runner == 'H100' && 'linux-amd64-gpu-h100-latest-1-testing') }}
     # Our self-hosted runners require a container
     # TODO: use a different (nvidia?) container
     container:
@@ -256,13 +263,6 @@ jobs:
         with:
           python-version: ${{ matrix.python-version }}
 
-      # The cache action needs this
-      - name: Install zstd
-        shell: bash --noprofile --norc -xeuo pipefail {0}
-        run: |
-          apt update
-          apt install zstd
-
       - name: Set up mini CTK
         uses: ./.github/actions/fetch_ctk
         continue-on-error: false