[Fix XPU CI][Inductor UT] Fix test cases broken by community. (#161142)

Fixes #161384, Fixes #161162, Fixes #160946, Fixes #160947, Fixes #160948

Pull Request resolved: https://github.com/pytorch/pytorch/pull/161142
Approved by: https://github.com/jansel

test/inductor/test_fxir_backend.py

@@ -599,6 +599,8 @@ class AOTFxirTestCase(InductorTestCase):
     device = GPU_TYPE
 
     def check(self, model, inp, dynamic_shapes=None, strict=False):
+        if self.device == "xpu":
+            raise unittest.SkipTest("The feature AOTFxir not currently ready for XPU")
         with torch.no_grad():
             ep = torch.export.export(
                 model, inp, dynamic_shapes=dynamic_shapes, strict=strict

test/inductor/test_max_autotune.py

@@ -700,6 +700,9 @@ class TestMaxAutotune(TestCase):
     @config.patch(max_autotune_gemm_backends="TRITON")
     @parametrize("search_space", ("DEFAULT", "EXHAUSTIVE"))
     def test_baddmm(self, search_space):
+        if search_space == "EXHAUSTIVE" and GPU_TYPE == "xpu":
+            raise unittest.SkipTest("EXHAUSTIVE search take too much time on XPU")
+
         class M(torch.nn.Module):
             def __init__(self):
                 super().__init__()
@@ -1940,6 +1943,8 @@ class TestMaxAutotuneSubproc(TestCase):
         """
         Make sure autotuning addmm in sub processes work without crashes.
         """
+        if search_space == "EXHAUSTIVE" and GPU_TYPE == "xpu":
+            raise unittest.SkipTest("EXHAUSTIVE search take too much time on XPU")
 
         torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
 

test/inductor/test_torchinductor_strided_blocks.py

@@ -381,11 +381,19 @@ class CommonTemplate:
         input_reader = InputReader()
         load_args(input_reader)
         args = input_reader.args
+        if self.device == "xpu":
+            atol = 1e-7
+            rtol = 1e-5
+        else:
+            atol = None
+            rtol = None
 
         self._run_and_compare(
             forward,
             *args,
             expected_num_block_pointers=4,
+            atol=atol,
+            rtol=rtol,
         )
 
     @parametrize(

test/run_test.py

@@ -250,6 +250,7 @@ XPU_BLOCKLIST = [
     "profiler/test_profiler_tree",
     "profiler/test_record_function",
     "profiler/test_torch_tidy",
+    "test_openreg",
 ]
 
 XPU_TEST = [

torch/_inductor/template_heuristics/triton.py

@@ -1351,6 +1351,13 @@ class XPUConfigHeuristic(BaseConfigHeuristic):
 
         return flex_decode_configs
 
+    def _prune_exhaustive_configs(
+        self,
+        configs: list[BaseConfig],
+        dtype_size: int,
+    ) -> list[BaseConfig]:
+        return configs
+
 
 class MTIAConfigHeuristic(BaseConfigHeuristic):
     """

torch/testing/_internal/common_methods_invocations.py

@@ -21078,6 +21078,7 @@ op_db: list[OpInfo] = [
             # NOTE: Only run on MPS
             DecorateInfo(unittest.skip('Skipped!'), device_type='cpu'),
             DecorateInfo(unittest.skip('Skipped!'), device_type='cuda'),
+            DecorateInfo(unittest.skip('Skipped!'), device_type='xpu'),
             DecorateInfo(unittest.skip('Skipped!'), device_type='meta'),
         ),),
     OpInfo(