[CI] Refactor CICD test workflows

.github/actions/get-runner/action.yml

@@ -0,0 +1,47 @@
+name: Get Runner Infos
+
+outputs:
+  runner_id:
+    value: ${{ steps.runner.outputs.runner_id }}
+  user_id:
+    value: ${{ steps.runner.outputs.user_id }}
+  render_id:
+    value: ${{ steps.runner.outputs.render_id }}
+  hostname:
+    value: ${{ steps.runner.outputs.hostname }}
+
+permissions: read-all
+
+runs:
+  using: composite
+  steps:
+    - name: Get runner
+      shell: bash -xe {0}
+      id: runner
+      run: |
+        # get test runner
+        echo "runner_id=$(echo ${RUNNER_NAME} |sed 's/\-[0-9]$//')" |tee -a ${GITHUB_OUTPUT}
+        echo "user_id=$(id -u)" |tee -a ${GITHUB_OUTPUT}
+        echo "render_id=$(getent group render |cut -d: -f3)" |tee -a ${GITHUB_OUTPUT}
+        echo "hostname=$(hostname)" |tee -a ${GITHUB_OUTPUT}
+        # show host info
+        lscpu
+        lshw -C display
+        free -h
+        df -h
+        cat /etc/os-release
+        uname -a
+    - name: Cleanup host
+      shell: bash -xe {0}
+      run: |
+        # clean docker cache
+        docker system prune -af || true
+        # clean workspace
+        ls -al
+        sudo find ./ |grep -v "^\./$" |xargs sudo rm -rf
+        cd ${RUNNER_WORKSPACE}/..
+        if [ "${PWD}" != "/" ];then
+          ls -al
+          sudo chmod 777 -R torch-xpu-ops _temp _actions _tool || true
+          sudo rm -rf _temp
+        fi

.github/actions/inductor-xpu-e2e-test/action.yml → .github/actions/linux-e2etest/action.yml

@@ -1,14 +1,14 @@
-name: inductor-xpu-e2e-test
+name: Linux E2E Test
 
 inputs:
+  env_prepare:
+    required: false
+    description: If set to any value, will prepare suite test env
   suite:
     required: true
     type: string
     default: 'huggingface'
     description: Dynamo benchmarks test suite. huggingface,timm_models,torchbench. Delimiter is comma
-  env_prepare:
-    required: false
-    description: If set to any value, will prepare suite test env
   dt:
     required: true
     type: string
@@ -24,49 +24,24 @@ inputs:
     type: string
     default: 'accuracy'
     description: accuracy,performance. Delimiter is comma
-  cards:
-    required: false
-    type: string
-    default: 'all'
-    description: which cards can be used in the test
-  hf_token:
-    required: false
-    description: HUGGING_FACE_HUB_TOKEN for torchbench test
-  pytorch:
-    required: false
-    type: string
-    default: 'main'
-    description: Pytorch branch/commit
-  driver:
-    required: false
-    type: string
-    default: 'lts'
-    description: Driver lts/rolling
 
 runs:
   using: composite
   steps:
     - name: Prepare ENV
       if: ${{ inputs.env_prepare }}
-      shell: bash
+      shell: bash -xe {0}
       run: |
-        source activate e2e_ci
         if [[ ${{ inputs.suite }} == *"torchbench"* ]]; then
-          if [ "${{ inputs.pytorch }}" != "nightly_wheel" ]; then
-            cd ../ && rm -rf audio && git clone --single-branch -b main https://github.com/pytorch/audio.git
-            cd audio && git checkout $TORCHAUDIO_COMMIT_ID
-            python setup.py bdist_wheel && pip uninstall torchaudio -y && pip install dist/*.whl
-            cd ../ && rm -rf vision && git clone --single-branch -b main https://github.com/pytorch/vision.git
-            cd vision && git checkout $TORCHVISION_COMMIT_ID
-            python setup.py bdist_wheel && pip uninstall torchvision -y && pip install dist/*.whl
-          fi
-          cd ../ && python -c "import torch, torchvision, torchaudio"
-          rm -rf benchmark && git clone https://github.com/pytorch/benchmark.git
-          cd benchmark && git checkout $TORCHBENCH_COMMIT_ID
+          python -c "import torch, torchvision, torchaudio"
+          cd ./pytorch
+          TORCHBENCH_COMMIT_ID=$(cat .github/ci_commit_pins/torchbench.txt)
+          git clone https://github.com/pytorch/benchmark.git xpu-benchmark
+          cd xpu-benchmark && git checkout $TORCHBENCH_COMMIT_ID
           # remove deps which will reinstall torch
           pip install --no-deps accelerate
-          pip install --no-deps git+https://github.com/huggingface/pytorch-image-models@$TIMM_COMMIT_ID
-          pip install $(curl -sSL https://raw.githubusercontent.com/huggingface/pytorch-image-models/$TIMM_COMMIT_ID/requirements.txt | grep -vE torch)
+          pip install --no-deps git+https://github.com/huggingface/pytorch-image-models@v1.0.14
+          pip install $(curl -sSL https://raw.githubusercontent.com/huggingface/pytorch-image-models/v1.0.14/requirements.txt | grep -vE torch)
           pip install -U transformers==4.44.2
           sed -i 's+.*pytorch-image-models.*++g;s+^accelerate.*++g;s/^transformers.*//g'  requirements.txt
           git status && git diff
@@ -81,28 +56,18 @@ runs:
           pip install -U transformers==4.44.2
         fi
         if [[ ${{ inputs.suite }} == *"timm_models"* ]]; then
-          if [ "${{ inputs.pytorch }}" != "nightly_wheel" ]; then
-            cd ../ && rm -rf vision && git clone --single-branch -b main https://github.com/pytorch/vision.git
-            cd vision && git checkout $TORCHVISION_COMMIT_ID
-            python setup.py bdist_wheel && pip uninstall torchvision -y && pip install dist/*.whl
-          fi
           # install timm without dependencies
-          pip install --no-deps git+https://github.com/huggingface/pytorch-image-models@$TIMM_COMMIT_ID
+          pip install --no-deps git+https://github.com/huggingface/pytorch-image-models@v1.0.14
           # install timm dependencies without torch and torchvision
-          pip install $(curl -sSL https://raw.githubusercontent.com/huggingface/pytorch-image-models/$TIMM_COMMIT_ID/requirements.txt | grep -vE torch)
+          pip install $(curl -sSL https://raw.githubusercontent.com/huggingface/pytorch-image-models/v1.0.14/requirements.txt | grep -vE torch)
         fi
-        pip install numpy==1.26.4
+        pip install -U numpy==1.26.4
+        pip list |grep -E 'intel|torch'
     - name: E2E Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
-      env:
-        HUGGING_FACE_HUB_TOKEN: ${{ inputs.hf_token }}
-        NEOReadDebugKeys: ${{ inputs.driver == 'rolling' && '1' || '0' }}
-        DisableScratchPages: ${{ inputs.driver == 'rolling' && '1' || '0' }}
-      shell: bash
+      shell: bash -xe {0}
       run: |
-        source activate e2e_ci
-        cp .github/scripts/inductor_xpu_test.sh ../pytorch
-        cd ../pytorch
-
+        cp ./.github/scripts/inductor_xpu_test.sh ./pytorch
+        cd ./pytorch
         # check param
         function contains() {
             contains_status="echo 'Start $2 ...'"
@@ -113,7 +78,6 @@ runs:
               contains_status="continue"
             }
         }
-        set -xe
         xpu_num=$(clinfo --list |awk 'BEGIN{gpu=0;}{if(gpu==1 && $0~/Platform/){gpu=0;}; if(gpu==1){print $0;}; if($0~/Platform.*Graphics/){gpu=1;}}' |wc -l)
         cores_per_instance="$(lscpu |grep -E 'Core\(s\) per socket:|Socket\(s\):' |awk -v i="${xpu_num}" 'BEGIN{sum=1}{sum*=$NF}END{print sum/i}')"
         export OMP_NUM_THREADS=${cores_per_instance}
@@ -160,24 +124,19 @@ runs:
         done
 
     - name: Summary E2E Test (${{ inputs.suite }} ${{ inputs.dt }} ${{ inputs.mode }} ${{ inputs.scenario }})
-      env:
-        HUGGING_FACE_HUB_TOKEN: ${{ inputs.hf_token }}
-      shell: bash
+      shell: bash -xe {0}
       run: |
-        cd ../pytorch
+        cd ./pytorch
         rm -f inductor_log/summary_accuracy.csv
         for var in $(find inductor_log/ -name "inductor_*_xpu_accuracy.csv")
         do
           sed -i "s/$/,$(basename $var)/" $var
           cat $var >> inductor_log/summary_accuracy.csv
         done
-
-        source activate e2e_ci
         cd ${{ github.workspace }}
-        cp .github/scripts/inductor_summary.py ../pytorch
-        cd ../pytorch
+        cp ./.github/scripts/inductor_summary.py ./pytorch
+        cd ./pytorch
         pip install styleFrame scipy pandas
-        set -xe
         dt=$(echo ${{ inputs.dt }} |sed 's/,/ /g')
         mode=$(echo ${{ inputs.mode }} |sed 's/,/ /g')
         suite=$(echo ${{ inputs.suite }} |sed 's/,/ /g')

.github/actions/linux-testenv/action.yml

@@ -0,0 +1,135 @@
+name: Setup Test Environment
+
+inputs:
+  test_type:
+    required: true
+    type: string
+    description: Test scope
+  pytorch:
+    type: string
+    default: 'main'
+    description: Pytorch main by default, or 'commit/branch', or 'repo@commit/repo@branch'
+  torch_xpu_ops:
+    type: string
+    default: 'main'
+    description: Torch-xpu-ops main by default, 'commit/branch', or 'repo@commit/repo@branch', or 'pinned' for pytorch pin
+  oneapi:
+    type: string
+    default: 'installed'
+    description: Installed oneAPI DLE on host by default, fill offline.sh url if needed
+  python:
+    type: string
+    default: '3.10'
+    description: Python version
+
+permissions: read-all
+
+runs:
+  using: composite
+  steps:
+    - name: Setup python-${{ inputs.python }}
+      uses: actions/setup-python@v5
+      with:
+        python-version: ${{ inputs.python }}
+    - name: Check runner
+      shell: bash -xe {0}
+      run: |
+        hostname && id
+        cat /etc/os-release
+        gcc -v && g++ -v
+        which python && python -V
+        which pip && pip list
+        pip install -U pip wheel setuptools
+        uname -a
+        dpkg -l |grep -E 'libigc-dev|libze-dev|level-zero-dev'
+        clinfo --list
+        cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor |sort |uniq -c
+        rm -rf ~/.triton /tmp/*inductor*
+        pip install pandas psutil scipy requests pytest-timeout
+    - name: Checkout torch-xpu-ops
+      uses: actions/checkout@v4
+      with:
+        path: torch-xpu-ops
+    - name: Install oneAPI DLE
+      shell: bash -xe {0}
+      if: ${{ inputs.oneapi != 'installed' }}
+      run: |
+        rm -rf ~/intel ~/.intel
+        wget -q -O oneapi.sh "${{ inputs.oneapi }}"
+        bash oneapi.sh -a -s --eula accept --action install --install-dir ${HOME}/intel/oneapi
+        echo "XPU_ONEAPI_PATH=${HOME}/intel/oneapi" >> ${GITHUB_ENV}
+        source ${HOME}/intel/oneapi/setvars.sh
+        sycl-ls && icpx -v
+    - name: Download Pytorch wheel
+      if: ${{ ! contains(inputs.test_type, 'wheel') }}
+      uses: actions/download-artifact@v4
+      with:
+        pattern: Torch-XPU-Wheel-*
+    - name: Prepare Stock Pytorch
+      shell: bash -xe {0}
+      run: |
+        # install pytorch
+        if [ $(echo "${{ inputs.pytorch }}" |grep -w "release_wheel" -c) -ne 0 ];then
+          pip install torch torchvision torchaudio --pre --index-url https://download.pytorch.org/whl/xpu
+        elif [ $(echo "${{ inputs.pytorch }}" |grep -w "test_wheel" -c) -ne 0 ];then
+          pip install torch torchvision torchaudio --pre --index-url https://download.pytorch.org/whl/test/xpu
+        elif [ $(echo "${{ inputs.pytorch }}" |grep -w "nightly_wheel" -c) -ne 0 ];then
+          pip install torch torchvision torchaudio --pre --index-url https://download.pytorch.org/whl/nightly/xpu
+        else
+          pip install --force-reinstall $(find ${{ github.workspace }}/ -name "*torch*.whl")
+        fi
+        pip list |grep torch
+        TORCH_COMMIT_ID=$(python -c 'import torch; print(torch.version.git_version)')
+        if [[ "${{ inputs.pytorch }}" == *"https://"* ]];then
+          PYTORCH_REPO="$(echo ${{ inputs.pytorch }} |sed 's/@.*//')"
+        else
+          PYTORCH_REPO="https://github.com/pytorch/pytorch.git"
+        fi
+        git clone ${PYTORCH_REPO} pytorch
+        cd pytorch
+        git checkout ${TORCH_COMMIT_ID}
+        pip install -r .ci/docker/requirements-ci.txt
+        # apply extra PRs for stock pytorch
+        if [[ "${{ inputs.test_type }}" == *"cicd"* ]];then
+          python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py -e https://github.com/pytorch/pytorch/pull/152940
+        else
+          python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
+        fi
+        git status && git diff && git show -s
+    - name: Prepare Torch-xpu-ops
+      shell: bash -xe {0}
+      if: ${{ inputs.torch_xpu_ops != 'skipped' }}
+      run: |
+        cd pytorch
+        rm -rf third_party/torch-xpu-ops
+        if [[ "${{ inputs.torch_xpu_ops }}" == *"https://"* ]];then
+          TORCH_XPU_OPS_REPO="$(echo ${{ inputs.torch_xpu_ops }} |sed 's/@.*//')"
+          TORCH_XPU_OPS_COMMIT="$(echo ${{ inputs.torch_xpu_ops }} |sed 's/.*@//')"
+        else
+          TORCH_XPU_OPS_REPO="https://github.com/intel/torch-xpu-ops.git"
+          if [ "${{ inputs.torch_xpu_ops }}" == "pinned" ];then
+            TORCH_XPU_OPS_COMMIT="$(cat third_party/xpu.txt)"
+          else
+            TORCH_XPU_OPS_COMMIT="${{ inputs.torch_xpu_ops }}"
+          fi
+        fi
+        if [ "${{ inputs.test_type }}" == "cicd" ];then
+          cp -r ${{ github.workspace }}/torch-xpu-ops third_party/torch-xpu-ops
+        else
+          git clone ${TORCH_XPU_OPS_REPO} third_party/torch-xpu-ops
+        fi
+        cd third_party/torch-xpu-ops
+        git checkout ${TORCH_XPU_OPS_COMMIT}
+        git status && git diff && git show -s
+    - name: Torch Config
+      shell: bash -xe {0}
+      run: |
+        printenv
+        python -c "import torch; print(torch.__config__.show())"
+        python -c "import torch; print(torch.__config__.parallel_info())"
+        python -c "import torch; print(torch.__config__.torch.xpu.device_count())"
+        python -c "import torchvision; print(torchvision.__version__)"
+        python -c "import torchaudio; print(torchaudio.__version__)"
+        python -c "import triton; print(triton.__version__)"
+        python pytorch/torch/utils/collect_env.py
+        pip list |grep -E 'torch|intel'