Route _neg_view_copy to ET neg kernel (pytorch#24194) (pytorch#24194)

Summary:

`aten::_neg_view_copy` (negation with view-copy autograd metadata,
identical numerics to neg for inference) had no provider route, crashing
at backend init (0x10). Add it to the ET spec table mapped to
`aten::neg.out`.

Reviewed By: digantdesai

Differential Revision: D126929421

Author: JakeStevens

backends/cpu/runtime/providers/reference/ReferenceProvider.cpp

@@ -106,6 +106,7 @@ constexpr auto kKernels = std::to_array<KernelSpec>({
      nullptr,
      false,
      "56"},
+    {"torch.ops.aten._neg_view_copy.default", "aten::neg.out", 1, 1, nullptr},
     {"torch.ops.aten._pdist_forward.default",
      "aten::_pdist_forward.out",
      2,
@@ -1021,6 +1022,20 @@ class ReferenceExecutable final : public Executable {
             reason.data());
         return Error::NotSupported;
       }
+      if (node.target == "torch.ops.aten._neg_view_copy.default" &&
+          tensor_bytes(context_.graph.value(node.inputs[0].arg.as_tensor().id))
+                  .get() == 0) {
+        const auto& input =
+            context_.graph.value(node.inputs[0].arg.as_tensor().id)
+                .tensor_meta();
+        const auto& output =
+            context_.graph.value(node.outputs[0].value_id).tensor_meta();
+        ET_CHECK_OR_RETURN_ERROR(
+            input.dtype == output.dtype && input.sizes == output.sizes,
+            InvalidProgram,
+            "Empty negative view copy requires matching input/output metadata");
+        continue;
+      }
       auto instruction = std::make_unique<Instruction>();
       instruction->node = &node;
       instruction->kernel = route->kernel;

backends/cpu/test/test_runtime.py

@@ -95,6 +95,29 @@ def forward(self, x, mask):
                 for output, expected in zip(actual, model(x, current), strict=True):
                     torch.testing.assert_close(output, expected, rtol=0, atol=0)
 
+    def test_neg_view_copy_routes_to_neg(self):
+        class NegView(torch.nn.Module):
+            def forward(self, x):
+                return torch.ops.aten._neg_view_copy.default(x)
+
+        model = NegView()
+        x = torch.randn(2, 3)
+        with self._lower(model, (x,)) as runtime:
+            actual = runtime.forward((x,))[0]
+            torch.testing.assert_close(actual, model(x))
+
+    def test_neg_view_copy_empty_bool(self):
+        class NegView(torch.nn.Module):
+            def forward(self, x):
+                return torch.ops.aten._neg_view_copy.default(x)
+
+        model = NegView()
+        x = torch.empty(0, dtype=torch.bool)
+        expected = model(x)
+        with self._lower(model, (x,)) as runtime:
+            actual = runtime.forward((x,))[0]
+            torch.testing.assert_close(actual, expected)
+
     def test_optional_integer_index_lists(self):
         class Index(torch.nn.Module):
             def __init__(self, accumulate):