use stock pytorch defintion of python_ref_db for test_ops (#2303)

disable_build
disable_e2e
disable_distributed

Author: daisyden

test/xpu/skip_list_common.py

@@ -14,6 +14,10 @@
         "test_python_ref_executor__refs_mul_executor_aten_xpu_complex32",
         # https://github.com/intel/torch-xpu-ops/issues/2254
         "histogramdd",
+        "_vdot_",
+        "_dot_",
+        "_flash_attention_",
+        "_efficient_attention_",
     ),
     "test_binary_ufuncs_xpu.py": (
         "test_fmod_remainder_by_zero_integral_xpu_int64",  # zero division is an undefined behavior: different handles on different backends

test/xpu/xpu_test_utils.py

@@ -7,7 +7,7 @@
 import unittest
 
 import torch
-from torch import bfloat16, cuda
+from torch import cuda
 from torch.testing._internal import (
     common_cuda,
     common_device_type,
@@ -354,6 +354,11 @@
         "_refs.div",
         "test_python_ref_torch_fallback",
     ),
+    ("_refs.true_div", "test_python_ref"),
+    (
+        "_refs.true_div",
+        "test_python_ref_torch_fallback",
+    ),
     ("argsort", "test_non_standard_bool_values"),
     ("sort", "test_non_standard_bool_values"),
 ]
@@ -865,7 +870,6 @@ def __init__(self, patch_test_case=True) -> None:
         )
         self.foreach_reduce_op_db = common_methods_invocations.foreach_reduce_op_db
         self.foreach_other_op_db = common_methods_invocations.foreach_other_op_db
-        self.python_ref_db = common_methods_invocations.python_ref_db
         self.ops_and_refs = common_methods_invocations.ops_and_refs
         self.largeTensorTest = common_device_type.largeTensorTest
         self.TEST_CUDA = common_cuda.TEST_CUDA
@@ -921,19 +925,10 @@ def gen_xpu_wrappers(op_name, wrappers):
 
     def align_supported_dtypes(self, db):
         for opinfo in db:
-            if (
-                opinfo.name not in _xpu_computation_op_list
-                and (
-                    opinfo.torch_opinfo.name not in _xpu_computation_op_list
-                    if db == common_methods_invocations.python_ref_db
-                    else True
-                )
-            ) or opinfo.name in _ops_without_cuda_support:
+            if opinfo.name in _ops_without_cuda_support:
                 opinfo.dtypesIf["xpu"] = opinfo.dtypes
             else:
                 backward_dtypes = set(opinfo.backward_dtypesIfCUDA)
-                if bfloat16 in opinfo.dtypesIf["xpu"]:
-                    backward_dtypes.add(bfloat16)
                 opinfo.backward_dtypes = tuple(backward_dtypes)
 
             if opinfo.name in _ops_dtype_different_cuda_support:
@@ -1039,13 +1034,13 @@ def __init__(self, *args):
             self.align_db_decorators(db)
             self.filter_fp64_sample_input(db)
         self.align_db_decorators(module_db)
-        common_methods_invocations.python_ref_db = [
+        _python_ref_db = [
             op
-            for op in self.python_ref_db
+            for op in common_methods_invocations.python_ref_db
             if op.torch_opinfo_name in _xpu_computation_op_list
         ]
         common_methods_invocations.ops_and_refs = (
-            common_methods_invocations.op_db + common_methods_invocations.python_ref_db
+            common_methods_invocations.op_db + _python_ref_db
         )
         common_methods_invocations.unary_ufuncs = [
             op
@@ -1128,7 +1123,6 @@ def __exit__(self, exc_type, exc_value, traceback):
             self.instantiate_parametrized_tests_fn
         )
         common_utils.TestCase = self.test_case_cls
-        common_methods_invocations.python_ref_db = self.python_ref_db
         common_methods_invocations.ops_and_refs = self.ops_and_refs
         common_device_type.largeTensorTest = self.largeTensorTest
         common_cuda.TEST_CUDA = self.TEST_CUDA