pytorch · Silv3S · May 14, 2025 · May 15, 2025 · May 15, 2025 · May 16, 2025
diff --git a/test/inductor/test_cpu_repro.py b/test/inductor/test_cpu_repro.py
@@ -5318,6 +5318,15 @@ def fn(x):
 
         self.common(fn, (x,))
 
+    def test_vector_norm_compile(self):
+        x = torch.randn([16, 32], dtype=torch.float)
+        ref = torch.linalg.vector_norm(x, ord=2, dim=[], keepdim=False, dtype=None)
+        compiled_vector_norm = torch.compile(
+            torch.linalg.vector_norm, backend="inductor"
+        )
+        res = compiled_vector_norm(x, ord=2, dim=[], keepdim=False, dtype=None)
+        self.assertEqual(ref, res)
+
 
 if __name__ == "__main__":
     from torch._inductor.test_case import run_tests

@@ -65,7 +65,7 @@ def _rand_shape(dim, min_size, max_size):
         shape.append(random.randint(min_size, max_size))
     return tuple(shape)
 
-def _reduced_shape(shape, dim=None, keepdim=False):
+def _reduced_shape(shape, empty_dim_as_none=False, dim=None, keepdim=False):
     """Computes the expected reduced shape given dim and keepdim
 
     Args:
@@ -77,7 +77,7 @@ def _reduced_shape(shape, dim=None, keepdim=False):
     Returns:
         The reduced shape
     """
-    if dim is None:
+    if dim is None or (empty_dim_as_none and dim == []):
         return [1] * len(shape) if keepdim else []
 
     # Wrap negative dims
@@ -105,7 +105,8 @@ def _test_dim_keepdim(self, op: ReductionOpInfo, device, *, ndim, **dim_keepdim)
         t = make_tensor(shape, dtype=torch.float, device=device)
         args, kwargs = next(op.generate_args_kwargs(t, **dim_keepdim))
         result = op(t, *args, **dim_keepdim, **kwargs)
-        expected_shape = _reduced_shape(shape, **dim_keepdim)
+        empty_dim_as_none = (op.name == "linalg.vector_norm" or op.name == "_refs.linalg.vector_norm")
+        expected_shape = _reduced_shape(shape, empty_dim_as_none, **dim_keepdim)
         self.assertEqual(result.shape, expected_shape, f"""
         expected output shape to be {expected_shape} but got {list(result.shape)}
         for input shape {shape} and {dim_keepdim}
@@ -314,7 +315,7 @@ def test_empty_tensor_nonempty_slice(self, device, op: ReductionOpInfo):
         for dim in [1] + [[1, 2]] if op.supports_multiple_dims else []:
             args, kwargs = next(op.generate_args_kwargs(t, dim=dim))
             result = op(t, *args, dim=dim, **kwargs)
-            self.assertEqual(result.shape, _reduced_shape(t.shape, dim))
+            self.assertEqual(result.shape, _reduced_shape(t.shape, dim=dim))
 
     def _test_noncontiguous(self, op: ReductionOpInfo, t: torch.Tensor, **reduction_kwargs):
         """Helper method to test noncontiguous input tensors."""

diff --git a/torch/_refs/linalg/__init__.py b/torch/_refs/linalg/__init__.py
@@ -165,6 +165,9 @@ def vector_norm(
         reduce_sum = partial(torch.sum, dim=dim, keepdim=keepdim)
 
         is_ord_even = ord % 2 == 0 if isinstance(ord, IntLike) else ord % 2.0 == 0.0
+        if dim == []:
+            dim = None
+
         if (dim is None and x.numel() == 1) or (
             dim is not None
             and (x.ndim > 0 and all(guard_size_oblivious(x.shape[d] == 1) for d in dim))

diff --git a/torch/testing/_internal/opinfo/definitions/linalg.py b/torch/testing/_internal/opinfo/definitions/linalg.py
@@ -1751,13 +1751,6 @@ def make_input():
         dtypes=floating_and_complex_types_and(torch.float16, torch.bfloat16),
         generate_args_kwargs=sample_kwargs_vector_norm,
         aten_name="linalg_vector_norm",
-        skips=(
-            # FIXME: sum reduces all dimensions when dim=[]
-            DecorateInfo(unittest.expectedFailure, "TestReductions", "test_dim_empty"),
-            DecorateInfo(
-                unittest.expectedFailure, "TestReductions", "test_dim_empty_keepdim"
-            ),
-        ),
     ),
     OpInfo(
         "linalg.lu_factor",