[CI] Add xpu test for windows

.github/workflows/_windows_ut.yml

@@ -150,3 +150,13 @@ jobs:
           set PYTORCH_TEST_WITH_SLOW=1
           cd ../pytorch/third_party/torch-xpu-ops/test/xpu/extended/
           python run_test_with_skip_mtl.py
+
+      - name: Run Test XPU UT
+        if: contains(inputs.ut, 'torch_xpu') || github.event_name == 'schedule'
+        shell: cmd
+        run: |
+          call "C:\ProgramData\miniforge3\Scripts\activate.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

.github/workflows/pull.yml

@@ -240,5 +240,5 @@ jobs:
     needs: preci-lint-check
     uses: ./.github/workflows/_windows_ut.yml
     with: 
-      ut: op_extended
+      ut: op_extended,torch_xpu
       runner: Windows_CI

test/xpu/run_test_win_with_skip_mtl.py

@@ -0,0 +1,17 @@
+import sys
+
+import pytest
+from skip_list_win_mtl import skip_dict
+
+IS_WINDOWS = sys.platform == "win32"
+
+skip_list = skip_dict["test_xpu_xpu.py"]
+
+skip_options = "not " + skip_list[0]
+for skip_case in skip_list[1:]:
+    skip_option = " and not " + skip_case
+    skip_options += skip_option
+
+test_command = ["-k", skip_options, "test_xpu_xpu.py", "-v"]
+res = pytest.main(test_command)
+sys.exit(res)

test/xpu/skip_list_win_mtl.py

@@ -0,0 +1,8 @@
+skip_dict = {
+    # failed on MTL windows, skip first for Preci
+    "test_xpu_xpu.py": (
+        "test_lazy_init_xpu",
+        "test_mem_get_info_xpu",
+        "test_wrong_xpu_fork_xpu",
+    ),
+}

test/xpu/test_xpu_xpu.py

@@ -0,0 +1,18 @@
+# Owner(s): ["module: intel"]
+
+from torch.testing._internal.common_device_type import instantiate_device_type_tests
+from torch.testing._internal.common_utils import run_tests, TestCase
+
+try:
+    from xpu_test_utils import XPUPatchForImport
+except Exception as e:
+    from .xpu_test_utils import XPUPatchForImport
+
+with XPUPatchForImport(False):
+    from test_xpu import TestXpu, TestXpuAutocast, TestXpuTrace  # noqa: F401
+
+instantiate_device_type_tests(TestXpu, globals(), only_for="xpu", allow_xpu=True)
+
+if __name__ == "__main__":
+    TestCase._default_dtype_check_enabled = True
+    run_tests()