Avoid COW ABI dependency in ATen bridge (#20808)

Summary:
- Add a local helper to fetch the mutable ATen tensor storage pointer
without calling Tensor::mutable_data_ptr().
- Use it when aliasing ATen tensors into ExecuTorch tensors so rebuilt
training pybindings do not depend on C10 COW materializer symbols.
- Preserve tensor-view offsets by applying storage_offset() *
itemsize().
- Import the XOR model relatively from the XOR exporter, so Android
source-tree model export does not load the stale installed training
pybinding before the wheel is rebuilt.

Why:
- _training_lib can fail to import when built against a Torch/C10 ABI
exposing materialize_cow_storage but loaded with a newer Torch runtime
where that symbol changed.
- The Android CI job installs the currently published executorch wheel,
so it needs the XOR exporter to avoid importing that stale native
training binding while generating test resources from source.

Author: JacobSzwejbka

extension/aten_util/aten_bridge.cpp

@@ -17,6 +17,22 @@ namespace executorch {
 namespace extension {
 
 namespace {
+void* mutable_tensor_data_ptr_no_cow(at::Tensor& tensor) {
+  if (tensor.numel() == 0) {
+    return nullptr;
+  }
+
+  auto* storage_data = static_cast<char*>(tensor.unsafeGetTensorImpl()
+                                              ->unsafe_storage()
+                                              .unsafeGetStorageImpl()
+                                              ->_mutable_data_ptr_no_checks()
+                                              .mutable_get());
+  ET_CHECK_MSG(
+      storage_data != nullptr,
+      "Tensor has a non-zero number of elements, but its data is not allocated");
+  return storage_data + tensor.storage_offset() * tensor.itemsize();
+}
+
 void check_tensor_meta(const at::Tensor& a, const executorch::aten::Tensor& b) {
   // 0-dim (scalar) tensors legitimately have empty sizes/strides arrays;
   // their `.data()` may return nullptr depending on the underlying container.
@@ -139,7 +155,8 @@ void alias_etensor_to_attensor(
       "Input tensor must have contiguous or channels last memory format");
 
   check_tensor_meta(aten_tensor, mutable_et);
-  mutable_et.unsafeGetTensorImpl()->set_data(aten_tensor.mutable_data_ptr());
+  mutable_et.unsafeGetTensorImpl()->set_data(
+      mutable_tensor_data_ptr_no_cow(aten_tensor));
 }
 
 at::Tensor alias_attensor_to_etensor(const torch::executor::Tensor& etensor) {
@@ -163,7 +180,7 @@ at::Tensor alias_attensor_to_etensor(const torch::executor::Tensor& etensor) {
 TensorPtr alias_tensor_ptr_to_attensor(at::Tensor& t) {
   return make_tensor_ptr(
       {t.sizes().begin(), t.sizes().end()},
-      t.mutable_data_ptr(),
+      mutable_tensor_data_ptr_no_cow(t),
       torch::executor::ScalarType(t.scalar_type()));
 }
 

extension/training/examples/XOR/export_model.py

@@ -13,7 +13,10 @@
 import torch
 from executorch.exir import ExecutorchBackendConfig, to_edge
 
-from executorch.extension.training.examples.XOR.model import Net, TrainingNet
+if __package__:
+    from .model import Net, TrainingNet
+else:
+    from model import Net, TrainingNet
 from torch.export import export
 from torch.export.experimental import _export_forward_backward