[Fix XPU CI][Inductor UT] Fix test cases broken by community. (#160403)

Fixes #160243, Fixes #160244, Fixes #160245

Pull Request resolved: https://github.com/pytorch/pytorch/pull/160403
Approved by: https://github.com/janeyx99

test/inductor/test_torchinductor_opinfo.py

@@ -26,6 +26,7 @@ from torch.testing._internal.common_device_type import (
     OpDTypes,
     ops,
     skipCPUIf,
+    skipCUDAIf,
     skipXPUIf,
 )
 from torch.testing._internal.common_methods_invocations import op_db, skipOps
@@ -45,11 +46,11 @@ from torch.testing._internal.common_utils import (
 from torch.testing._internal.inductor_utils import (
     GPU_TYPE,
     HAS_CPU,
+    HAS_CUDA_AND_TRITON,
     has_triton,
     HAS_XPU_AND_TRITON,
     maybe_skip_size_asserts,
 )
-from torch.testing._internal.triton_utils import requires_cuda_and_triton
 from torch.utils._dtype_abbrs import dtype_abbrs
 from torch.utils._python_dispatch import TorchDispatchMode
 from torch.utils._pytree import tree_map
@@ -682,6 +683,14 @@ inductor_override_kwargs["xpu"] = {
     ("nn.functional.unfold", f16): {
         "reference_in_float": True,
     },
+    # Reference crash on Intel LTS2 driver.
+    ("nn.functional.interpolate.trilinear", f32): {
+        "check_gradient": False,
+    },
+    # Reference crash on Intel LTS2 driver.
+    ("nn.functional.interpolate.trilinear", f64): {
+        "check_gradient": False,
+    },
 }
 if TEST_WITH_ROCM:
     inductor_override_kwargs["cuda"].update(
@@ -1125,7 +1134,7 @@ class TestInductorOpInfo(TestCase):
     @skipCUDAMemoryLeakCheckIf(
         True
     )  # inductor kernels failing this test intermittently
-    @requires_cuda_and_triton
+    @skipCUDAIf(not HAS_CUDA_AND_TRITON, "Skipped! Triton not found")
     @skipXPUIf(
         not HAS_XPU_AND_TRITON, "Skipped! Supported XPU compiler and Triton not found"
     )

torch/_inductor/ir.py

@@ -7078,10 +7078,10 @@ class DeviceCopy(ExternKernelOut):
             # x.get_stride() may be unimplemented if x's size is empty
             stride = x.get_stride()
         is_destination_pinned = (
-            x_device.type == "cuda" and device.type == "cpu" and non_blocking
+            is_gpu(x_device.type) and device.type == "cpu" and non_blocking
         )
         is_source_pinned = (
-            x_device.type == "cpu" and device.type == "cuda" and non_blocking
+            x_device.type == "cpu" and is_gpu(device.type) and non_blocking
         )
         if is_source_pinned and is_storage_and_layout(x):
             x.get_layout().is_pinned = True