Revert "fix Invalid call to aoti_torch_tensor_copy_ #123039 (#124037)"

This reverts commit 6e24cc0.

Reverted #124037 on behalf of https://github.com/jeanschmidt due to seems to have introduced a regression in pull / linux-focal-cuda12.1-py3.10-gcc9 / test (default, 3, 5, linux.4xlarge.nvidia.gpu) ([comment](#124037 (comment)))

test/inductor/test_cuda_cpp_wrapper.py

@@ -109,7 +109,9 @@ class DynamicShapesCudaWrapperCudaTests(InductorTestCase):
         test_failures_cuda_wrapper[
             f"{test_name}_dynamic_shapes"
         ] = test_torchinductor.TestFailure(("cuda_wrapper",), is_skip=False)
-    skip_list = []
+    skip_list = [
+        "test_multi_device_cuda",
+    ]
     for test_name in skip_list:
         test_failures_cuda_wrapper[test_name] = test_torchinductor.TestFailure(
             ("cuda_wrapper",), is_skip=True

torch/_inductor/codegen/cpp_wrapper_cpu.py

@@ -895,11 +895,9 @@ def codegen_scalar_to_tensor(self, output: str):
     @cache_on_self
     def get_output_refs(self):
         return [
-            (
-                f"torch::tensor({x.codegen_reference(self.wrapper_call)})"
-                if isinstance(x, ir.ShapeAsConstantBuffer) and not config.abi_compatible
-                else x.codegen_reference(self.wrapper_call)
-            )
+            f"torch::tensor({x.codegen_reference(self.wrapper_call)})"
+            if isinstance(x, ir.ShapeAsConstantBuffer) and not config.abi_compatible
+            else x.codegen_reference(self.wrapper_call)
             for x in V.graph.graph_outputs
         ]
 
@@ -1099,11 +1097,9 @@ def generate_end(self, result):
             outputs_str = "output_tensors"
         else:
             outputs = [
-                (
-                    f"output_tensors[{i}]"
-                    if self.output_is_tensor[i]
-                    else f"output_tensors[{i}].item()"
-                )
+                f"output_tensors[{i}]"
+                if self.output_is_tensor[i]
+                else f"output_tensors[{i}].item()"
                 for i in range(len(V.graph.graph_outputs))
             ]
             outputs_str = f"[{', '.join(outputs)}]"
@@ -1398,7 +1394,6 @@ def can_stack_allocate_buffer(self, buffer):
             and ir.is_contiguous_strides_for_shape(
                 buffer.get_stride(), buffer.get_size()
             )
-            and not buffer.is_extern()
         )
 
     def make_buffer_free(self, buffer):