Optimize preci & ngihtly tests (#1162)

1. Additional PRs apply script for windows 2. Checkout the corresponding torch & torch-xpu-ops versions for wheel tests 3. Don't need source DL-Essential for nightly wheel tests
intel · Dec 12, 2024 · 164c629 · 164c629
1 parent 66ac930
commit 164c629
Show file tree

Hide file tree

Showing 5 changed files with 48 additions and 32 deletions.
diff --git a/.github/actions/inductor-xpu-e2e-test/action.yml b/.github/actions/inductor-xpu-e2e-test/action.yml
@@ -51,7 +51,7 @@ runs:
       shell: bash
       run: |
         source activate e2e_ci
-        source .github/scripts/env.sh
+        source .github/scripts/env.sh ${{ inputs.pytorch }}
         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
@@ -94,7 +94,7 @@ runs:
       shell: bash
       run: |
         source activate e2e_ci
-        source .github/scripts/env.sh
+        source .github/scripts/env.sh ${{ inputs.pytorch }}
         cp .github/scripts/inductor_xpu_test.sh ../pytorch
         cd ../pytorch
 

diff --git a/.github/scripts/apply_torch_pr.py b/.github/scripts/apply_torch_pr.py
@@ -58,7 +58,7 @@ def appyly_pr(pr_info, re_apply_msg):
     pr_file = pr_info["diff_url"].split("/")[-1]
     urllib.request.urlretrieve(pr_info["diff_url"], pr_file)
     # apply diff
-    apply_cmd = "git apply --3way " + pr_file + " && rm -f " + pr_file
+    apply_cmd = "git apply --3way " + pr_file
     apply_info = subprocess.Popen(apply_cmd, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, shell=True)
     apply_message = apply_info.communicate()[0].decode("utf-8")
     apply_status = apply_info.returncode

diff --git a/.github/scripts/env.sh b/.github/scripts/env.sh
@@ -1,4 +1,9 @@
 #!/bin/bash
-source /opt/intel/oneapi/compiler/latest/env/vars.sh
-source /opt/intel/oneapi/umf/latest/env/vars.sh
-source /opt/intel/oneapi/pti/latest/env/vars.sh
+
+if [ "$1" != "nightly_wheel" ];then
+    source /opt/intel/oneapi/compiler/latest/env/vars.sh
+    source /opt/intel/oneapi/umf/latest/env/vars.sh
+    source /opt/intel/oneapi/pti/latest/env/vars.sh
+else
+    echo "Don't need to source DL-Essential for nightly wheel"
+fi
diff --git a/.github/workflows/_linux_ut.yml b/.github/workflows/_linux_ut.yml
@@ -65,19 +65,21 @@ jobs:
           conda create -n xpu_op_${ZE_AFFINITY_MASK} python=${{ inputs.python }} cmake ninja -y
           source activate xpu_op_${ZE_AFFINITY_MASK}
           cd ../ && rm -rf pytorch
-          git clone https://github.com/pytorch/pytorch pytorch
-          cd pytorch && git checkout $(echo ${{ inputs.pytorch }} |sed 's/^nightly_wheel$/nightly/')
-          # apply PRs for stock pytorch
           pip install requests
-          python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
-          git status && git show -s
-          git submodule sync && git submodule update --init --recursive
-          if [[ ${{ inputs.keep_torch_xpu_ops }} == 'true' ]]; then
-            echo "Don't replace torch-xpu-ops!"
-          else
-            rm -rf third_party/torch-xpu-ops && cp -r ../torch-xpu-ops third_party/
-            # Workaround for torch-xpu-ops ci test
-            sed -i "s/checkout --quiet \${TORCH_XPU_OPS_COMMIT}/log -n 1/g" caffe2/CMakeLists.txt
+          git clone https://github.com/pytorch/pytorch pytorch
+          if [ "${{ inputs.pytorch }}" != "nightly_wheel" ]; then
+            cd pytorch && git checkout $(echo ${{ inputs.pytorch }})
+            # apply PRs for stock pytorch
+            python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
+            git status && git show -s
+            git submodule sync && git submodule update --init --recursive
+            if [[ ${{ inputs.keep_torch_xpu_ops }} == 'true' ]]; then
+              echo "Don't replace torch-xpu-ops!"
+            else
+              rm -rf third_party/torch-xpu-ops && cp -r ../torch-xpu-ops third_party/
+              # Workaround for torch-xpu-ops ci test
+              sed -i "s/checkout --quiet \${TORCH_XPU_OPS_COMMIT}/log -n 1/g" caffe2/CMakeLists.txt
+            fi
           fi
       - name: Triton Installation
         run: |
@@ -96,28 +98,37 @@ jobs:
       - name: Build Pytorch XPU
         run: |
           source activate xpu_op_${ZE_AFFINITY_MASK}
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           pip install mkl-static==2025.0.1 mkl-include==2025.0.1
-          cd ../pytorch
           if [[ ${{ inputs.abi }} == '0' ]]; then
             export _GLIBCXX_USE_CXX11_ABI=0
           else
             export _GLIBCXX_USE_CXX11_ABI=1
           fi
           if [ "${{ inputs.pytorch }}" != "nightly_wheel" ]; then
+            cd ../pytorch
             export CMAKE_PREFIX_PATH=${CMAKE_PREFIX_PATH}:${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
             pip install -r requirements.txt
             WERROR=1 python setup.py bdist_wheel
             pip install --force-reinstall dist/*.whl
             git clone https://github.com/pytorch/vision && cd vision && python setup.py install && cd ..
           else
             pip install torch torchvision torchaudio --pre --index-url https://download.pytorch.org/whl/nightly/xpu
+            TORCH_COMMIT_ID=$(python -c 'import torch; print(torch.version.git_version)')
+            cd ../pytorch
+            git reset --hard && git checkout ${TORCH_COMMIT_ID}
+            TORCH_XPU_OPS_COMMIT=$(<third_party/xpu.txt)
+            rm -rf third_party/torch-xpu-ops
+            git clone https://github.com/intel/torch-xpu-ops.git third_party/torch-xpu-ops
+            cd third_party/torch-xpu-ops
+            git checkout ${TORCH_XPU_OPS_COMMIT}
+            cd ../..
           fi
           pip install -r .ci/docker/requirements-ci.txt
       - name: Torch Config
         run: |
           source activate xpu_op_${ZE_AFFINITY_MASK}
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           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())"
@@ -132,7 +143,7 @@ jobs:
         run: |
           cd ${{ github.workspace }}
           xpu-smi discovery
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           source activate xpu_op_${ZE_AFFINITY_MASK}
           cd ${{ github.workspace }}
           cd test/regressions
@@ -143,7 +154,7 @@ jobs:
         run: |
           cd ${{ github.workspace }}
           xpu-smi discovery
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           source activate xpu_op_${ZE_AFFINITY_MASK}
           export ZE_AFFINITY_MASK_OLD=${ZE_AFFINITY_MASK}
           unset ZE_AFFINITY_MASK
@@ -155,15 +166,15 @@ jobs:
       - name: Run XPU OP Extended UT
         if: contains(inputs.ut, 'op_extended') || github.event_name == 'schedule'
         run: |
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           source activate xpu_op_${ZE_AFFINITY_MASK}
           export PYTORCH_TEST_WITH_SLOW=1
           cd ../pytorch/third_party/torch-xpu-ops/test/xpu/extended/
           timeout 10000 python run_test_with_skip.py
       - name: Run XPU OP UT
         if: contains(inputs.ut, 'op_ut') || github.event_name == 'schedule'
         run: |
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           source activate xpu_op_${ZE_AFFINITY_MASK}
           export PYTORCH_ENABLE_XPU_FALLBACK=1
           export PYTORCH_TEST_WITH_SLOW=1
@@ -177,7 +188,7 @@ jobs:
       - name: Run Torch XPU UT
         if: contains(inputs.ut, 'torch_xpu') || github.event_name == 'schedule'
         run: |
-          source .github/scripts/env.sh
+          source .github/scripts/env.sh ${{ inputs.pytorch }}
           source activate xpu_op_${ZE_AFFINITY_MASK}
           cd ../pytorch
           TEST_REPORTS_DIR=$(pwd)/test/test-reports

diff --git a/.github/workflows/nightly_ondemand_whl.yml b/.github/workflows/nightly_ondemand_whl.yml
@@ -108,17 +108,18 @@ jobs:
           source activate e2e_ci
           pip install torch torchvision torchaudio --pre --index-url https://download.pytorch.org/whl/nightly/xpu
           echo "TORCH_BRANCH_ID=$(python -c 'import torch; print(torch.__version__)')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
-          echo "TORCH_COMMIT_ID=$(python -c 'import torch; print(torch.version.git_version)')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
-          source .github/scripts/env.sh
+          TORCH_COMMIT_ID=$(python -c 'import torch; print(torch.version.git_version)')
+          echo "TORCH_COMMIT_ID=${TORCH_COMMIT_ID}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           cd ../ && rm -rf pytorch
           git clone https://github.com/pytorch/pytorch pytorch
           cd pytorch && git checkout ${TORCH_COMMIT_ID}
           # apply PRs for stock pytorch
           pip install requests
-          python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
+          # python ../torch-xpu-ops/.github/scripts/apply_torch_pr.py
           git status && git show -s
           pip install -r requirements.txt
-          echo "TORCH_XPU_OPS_COMMIT=$(<third_party/xpu.txt)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          TORCH_XPU_OPS_COMMIT=$(<third_party/xpu.txt)
+          echo "TORCH_XPU_OPS_COMMIT=${TORCH_XPU_OPS_COMMIT}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           rm -rf third_party/torch-xpu-ops
           git clone https://github.com/intel/torch-xpu-ops.git third_party/torch-xpu-ops
           cd third_party/torch-xpu-ops
@@ -127,7 +128,6 @@ jobs:
         id: pinned
         run: |
           source activate e2e_ci
-          source .github/scripts/env.sh
           echo "TORCHVISION_COMMIT_ID=$(python -c 'import torchvision; print(torchvision.version.git_version)')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo "TORCHAUDIO_COMMIT_ID=$(python -c 'import torchaudio; print(torchaudio.version.git_version)')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo "TRITON_COMMIT_ID=$(python -c 'import triton; print(triton.__version__)')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
@@ -138,7 +138,7 @@ jobs:
           echo "MODEL_ONLY_NAME=${{ inputs.model }}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo "DRIVER_VERSION=$(dkms status 2>&1 |grep 'intel-i915-dkms' |sed 's/.*\///;s/,.*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo "KERNEL_VERSION=$(uname -rv 2>&1)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
-          echo "BUNDLE_VERSION=$(dpcpp --version 2>&1 |grep 'DPC++/C++' |sed 's/.*(//;s/).*//')" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
+          echo "BUNDLE_VERSION=$(pip list |grep cmplr |head -n 1)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           . /etc/os-release
           echo "OS_PRETTY_NAME=${PRETTY_NAME}" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"
           echo "GCC_VERSION=$(gcc -dumpversion)" |tee -a "${GITHUB_OUTPUT}" >> "${GITHUB_ENV}"