Merge branch 'main' into xiangdong/dist_cases

zxd1997066 · web-flow · commit f6588df8dd02 · 2025-09-11T10:56:30.000+08:00
diff --git a/.github/scripts/check-ut.py b/.github/scripts/check-ut.py
@@ -138,7 +138,6 @@ def print_failures(failure_list=None):
 
 def generate_failures_log():
     if not failures:
-        print("No failures found, skipping log file creation.")
         return
 
     for case in failures:
@@ -245,6 +244,8 @@ def determine_category(ut):
         return 'op_extended'
     elif ut == 'op_transformers':
         return 'op_transformers'
+    elif ut == 'test_xpu':
+        return 'test_xpu'
     elif 'op_ut' in ut:
         return 'op_ut'
     else:
@@ -296,7 +297,6 @@ def process_xml_file(xml_file):
 
 def generate_passed_log():
     if not passed_cases:
-        print("No passed cases found, skipping log file creation.")
         return
 
     for category, category_passed in passed_by_category.items():
diff --git a/.github/scripts/ut_result_check.sh b/.github/scripts/ut_result_check.sh
@@ -25,6 +25,10 @@ check_new_failed() {
 
     # Filter the same content from file_UT as file_known_issue
     echo "Filtering $file_known_issue for $file_UT"
+    if grep -q $'\r' "$file_UT"; then
+        echo "Detected log from windows"
+        sed -i 's/\r$//' "$file_UT"
+    fi
     grep -vFxf "$file_known_issue" "$file_UT" > "$output_file"
 
     echo -e "\n\033[1;31m[New failed cases Summary]\033[0m"
@@ -54,6 +58,10 @@ check_passed_known_issues() {
         return 1
     fi
     echo "Checking for known issues that are now passing in $file_passed_UT"
+    if grep -q $'\r' "$file_passed_UT"; then
+        echo "Detected log from windows"
+        sed -i 's/\r$//' "$file_passed_UT"
+    fi
     grep -Fxf "$file_passed_UT" "$file_known_issue" > "$output_file"
     echo -e "\n\033[1;32m[New passed cases Summary]\033[0m"
     if [[ -s "$output_file" ]]; then
@@ -74,6 +82,7 @@ check_test_cases() {
         ["op_regression_dev1"]=1
         ["op_transformers"]=237
         ["op_ut"]=120408
+        ["test_xpu"]=69
     )
 
     if [[ ! -f "$log_file" ]]; then
@@ -115,7 +124,7 @@ check_test_cases() {
 }
 
 
-if [[ "${ut_suite}" == 'op_regression' || "${ut_suite}" == 'op_regression_dev1' || "${ut_suite}" == 'op_extended' || "${ut_suite}" == 'op_transformers' || "${ut_suite}" == 'op_ut' ]]; then
+if [[ "${ut_suite}" == 'op_regression' || "${ut_suite}" == 'op_regression_dev1' || "${ut_suite}" == 'op_extended' || "${ut_suite}" == 'op_transformers' || "${ut_suite}" == 'op_ut' || "${ut_suite}" == 'test_xpu' ]]; then
     echo -e "========================================================================="
     echo -e "Show Failed cases in ${ut_suite}"
     echo -e "========================================================================="
@@ -188,35 +197,6 @@ if [[ "${ut_suite}" == 'op_regression' || "${ut_suite}" == 'op_regression_dev1'
     fi
 fi
 
-if [[ "${ut_suite}" == 'torch_xpu' ]]; then
-    echo "Pytorch XPU binary UT checking"
-    cd ../../pytorch || exit
-    for xpu_case in build/bin/*{xpu,sycl}*; do
-      if [[ "$xpu_case" != *"*"* && "$xpu_case" != *.so && "$xpu_case" != *.a ]]; then
-        case_name=$(basename "$xpu_case")
-        cd ../ut_log/torch_xpu || exit
-        grep -E "FAILED" binary_ut_"${ut_suite}"_"${case_name}"_test.log | awk '{print $2}' > ./binary_ut_"${ut_suite}"_"${case_name}"_failed.log
-        wc -l < "./binary_ut_${ut_suite}_${case_name}_failed.log" | tee -a ./binary_ut_"${ut_suite}"_failed_summary.log
-        grep -E "PASSED|Pass" binary_ut_"${ut_suite}"_"${case_name}"_test.log | awk '{print $2}' > ./binary_ut_"${ut_suite}"_"${case_name}"_passed.log
-        wc -l < "./binary_ut_${ut_suite}_${case_name}_passed.log" | tee -a ./binary_ut_"${ut_suite}"_passed_summary.log
-        cd - || exit
-      fi
-    done
-    echo -e "========================================================================="
-    echo -e "Show Failed cases in ${ut_suite}"
-    echo -e "========================================================================="
-    cd ../ut_log/torch_xpu || exit
-    cat "./binary_ut_${ut_suite}_${case_name}_failed.log"
-    num_failed_binary_ut=$(awk '{sum += $1};END {print sum}' binary_ut_"${ut_suite}"_failed_summary.log)
-    num_passed_binary_ut=$(awk '{sum += $1};END {print sum}' binary_ut_"${ut_suite}"_passed_summary.log)
-    ((num_failed=num_failed_binary_ut))
-    if [[ $num_failed -gt 0 ]] || [[ $num_passed_binary_ut -le 0 ]]; then
-      echo -e "[ERROR] UT ${ut_suite} test Fail"
-      exit 1
-    else
-      echo -e "[PASS] UT ${ut_suite} test Pass"
-    fi
-fi
 if [[ "${ut_suite}" == 'xpu_distributed' ]]; then
     grep -E "^FAILED" xpu_distributed_test.log | awk '{print $3}' > ./"${ut_suite}"_xpu_distributed_test_failed.log
     sed -i '/^[^.]\+/d' ./"${ut_suite}"_xpu_distributed_test_failed.log
diff --git a/.github/workflows/_windows_ut.yml b/.github/workflows/_windows_ut.yml
@@ -112,7 +112,7 @@ jobs:
           call conda activate windows_ci
           cd ../pytorch
           pip install -r requirements.txt
-          pip install cmake setuptools==72.1.0 clang-format
+          pip install cmake setuptools clang-format
           pip install mkl-static mkl-include
           set USE_STATIC_MKL=1
           copy "%CONDA_PREFIX%\Library\bin\libiomp*5md.dll" .\torch\lib
@@ -123,7 +123,8 @@ jobs:
               set CMAKE_PREFIX_PATH="%CONDA_PREFIX%\Library"
           )
           python setup.py clean
-          set MAX_JOBS=4
+          set MAX_JOBS=32
+          set TORCH_XPU_ARCH_LIST=mtl-h,bmg,lnl-m
           python setup.py bdist_wheel > build_torch_wheel_log.log
           echo "[INFO] begin to install torch whls"
           for /r C:\actions-runner\_work\torch-xpu-ops\pytorch\dist %%i in (torch*.whl) do (
@@ -141,24 +142,22 @@ jobs:
           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())"
-
       - name: Upload Windows build log
         if: ${{ ! cancelled() }}
         uses: actions/upload-artifact@v4
         with:
           name: Torch-XPU-Windows-Log-${{ github.event.pull_request.number || github.sha }}
           path: 'C:\actions-runner\_work\torch-xpu-ops\pytorch\build_torch_wheel_log.log'
-
       - name: Upload Windows binary
         if: ${{ ! cancelled() }}
         uses: actions/upload-artifact@v4
         with:
           name: Torch-XPU-Windows-Binary-${{ github.event.pull_request.number || github.sha }}
           path: 'C:\actions-runner\_work\torch-xpu-ops\pytorch\dist'
-
       - name: Run XPU OP Extended UT
         if: contains(inputs.ut, 'op_extended') || github.event_name == 'schedule'
         shell: cmd
+        continue-on-error: true
         run: |
           call "C:\ProgramData\miniforge3\Scripts\activate.bat"
           call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
@@ -168,13 +167,116 @@ jobs:
           cd ../pytorch/third_party/torch-xpu-ops/test/xpu/extended/
           python run_test_with_skip_mtl.py
 
+          if not exist "%GITHUB_WORKSPACE%\ut_log" mkdir "%GITHUB_WORKSPACE%\ut_log"
+          copy op_extended.xml %GITHUB_WORKSPACE%\ut_log /Y
       - name: Run Test XPU UT
-        if: contains(inputs.ut, 'torch_xpu') || github.event_name == 'schedule'
+        if: contains(inputs.ut, 'test_xpu') || github.event_name == 'schedule'
         shell: cmd
+        continue-on-error: true
         run: |
           call "C:\ProgramData\miniforge3\Scripts\activate.bat"
           call "C:\Program Files (x86)\Microsoft Visual Studio\2022\BuildTools\VC\Auxiliary\Build\vcvars64.bat"
           call conda activate windows_ci
           call "C:\Program Files (x86)\Intel\oneAPI\setvars.bat"
           cd ../pytorch/third_party/torch-xpu-ops/test/xpu/
           python run_test_win_with_skip_mtl.py
+
+          if not exist "%GITHUB_WORKSPACE%\ut_log" mkdir "%GITHUB_WORKSPACE%\ut_log"
+          copy test_xpu.xml %GITHUB_WORKSPACE%\ut_log /Y
+      - name: UT Test Results Summary
+        shell: cmd
+        run: |
+          call conda activate windows_ci
+          pip install junitparser
+          echo "GITHUB_WORKSPACE: %GITHUB_WORKSPACE%"
+          for %%i in ("%GITHUB_WORKSPACE%\ut_log\*.xml") do (
+            python .\.github\scripts\check-ut.py "%%i" >> "%GITHUB_STEP_SUMMARY%"
+          )
+          @echo off
+
+          REM Check the failure logs
+          if exist "%GITHUB_WORKSPACE%\failures*.log" (
+            echo Exist Failure logs
+            echo Found Failure logs as below: 
+            for %%f in ("%GITHUB_WORKSPACE%\failures*.log") do (
+              echo   - %%f
+              copy "%%f" "%GITHUB_WORKSPACE%\ut_log\"
+            )
+            echo Failure logs Copied
+          ) else (
+            echo No Failure logs
+          )
+
+          REM Copied the passed logs
+          if exist "passed*.log" (
+            copy "passed*.log" "%GITHUB_WORKSPACE%\ut_log\"
+            echo Passed logs Copied
+          ) else (
+            echo No Passed logs
+          )
+
+          REM Copied the Summary logs
+          if exist "category*.log" (
+            copy "category*.log" "%GITHUB_WORKSPACE%\ut_log\"
+            echo Category logs Copied
+          ) else (
+            echo No Category logs
+          )
+      - name: Upload Inductor XPU UT Log
+        if: ${{ ! cancelled() }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: Inductor-XPU-UT-Data-${{ github.event.pull_request.number || github.sha }}-Windows
+          path: "${{ github.workspace }}/ut_log"
+          if-no-files-found: ignore
+
+  summary:
+    needs: [ut_test]
+    runs-on: ubuntu-24.04
+    timeout-minutes: 30
+    env:
+      GH_TOKEN: ${{ github.token }}
+    steps:
+      - name: Checkout torch-xpu-ops
+        uses: actions/checkout@v4
+      - name: Download XPU UT Logs
+        uses: actions/download-artifact@v4
+        with:
+          name: Inductor-XPU-UT-Data-${{ github.event.pull_request.number || github.sha }}-Windows
+          path: ${{ github.workspace }}/ut_log
+      - name: Check UT Results
+        shell: bash
+        run: |
+          ls -al ${{ github.workspace }}/ut_log
+          cd ${{ github.workspace }}/ut_log
+
+          # get skipped known issues
+          count=$(gh api "repos/${{ github.repository }}/issues?labels=skipped_windows" --jq 'length')
+          if [ "$count" -gt 0 ]; then
+            echo -e "$count issues with skipped label found"
+            gh api "repos/${{ github.repository }}/issues?labels=skipped_windows" \
+              --jq '.[] | select(.pull_request == null) | "Issue #\(.number): \(.title)\n\(.body)\n"' > issues.log
+          fi
+
+          cp ${{ github.workspace }}/.github/scripts/ut_result_check.sh ./
+          for ut_name in $(echo ${{ inputs.ut }} |sed 's/,/ /g')
+          do
+            touch Known_issue.log
+            if [ -f "issues.log" ]; then
+              awk -v r="${ut_name}" 'BEGIN{ print_row = 0 }{
+                if ( ! ( $0 ~ /[a-zA-Z0-9]/ ) ) { print_row = 0 };
+                if ( print_row == 1 && $1 ~ r ) { print $0 };
+                if ( $0 ~ /Cases:/ ) { print_row = 1 };
+              }' issues.log > Known_issue.log
+            else
+              echo "Info: issues.log not found or empty, using empty Known_issue.log"
+            fi
+            bash ut_result_check.sh ${ut_name}
+          done
+      - name: Upload Inductor XPU UT Log
+        if: ${{ ! cancelled() }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: Inductor-XPU-UT-Data-${{ github.event.pull_request.number || github.sha }}-Windows
+          path: ${{ github.workspace }}/ut_log
+          overwrite: true
diff --git a/.github/workflows/nightly_ondemand.yml b/.github/workflows/nightly_ondemand.yml
@@ -209,7 +209,7 @@ jobs:
     secrets: inherit
     uses: ./.github/workflows/_windows_ut.yml
     with:
-      ut: 'op_extended,torch_xpu'
+      ut: 'op_extended,test_xpu'
       python: ${{ needs.Conditions-Filter.outputs.python }}
       src_changed: false
       has_label: true
diff --git a/.github/workflows/pull.yml b/.github/workflows/pull.yml
@@ -159,7 +159,7 @@ jobs:
     secrets: inherit
     uses: ./.github/workflows/_windows_ut.yml
     with: 
-      ut: op_extended,torch_xpu
+      ut: op_extended,test_xpu
       runner: Windows_CI
       src_changed: ${{ needs.conditions-filter.outputs.src_changed }}
       has_label: ${{ needs.conditions-filter.outputs.has_label }}
diff --git a/test/xpu/extended/run_test_with_skip_mtl.py b/test/xpu/extended/run_test_with_skip_mtl.py
@@ -19,5 +19,6 @@
 
 os.environ["PYTORCH_TEST_WITH_SLOW"] = "1"
 test_command = ["-k", skip_options, "test_ops_xpu.py", "-v"]
+test_command.extend(["--junit-xml", "./op_extended.xml"])
 res = pytest.main(test_command)
 sys.exit(res)
diff --git a/test/xpu/run_test_win_with_skip_mtl.py b/test/xpu/run_test_win_with_skip_mtl.py
@@ -18,6 +18,7 @@
 sys.stdout = StringIO()
 
 test_command = ["-k", skip_options, "../../../../test/test_xpu.py", "-v"]
+test_command.extend(["--junit-xml", "./test_xpu.xml"])
 res = pytest.main(test_command)
 
 output = sys.stdout.getvalue()
