EagerTensor_numpy now returns a view for supported dtypes

Two notes:

* Prior to this change it always produced a copy. See comment in
  EagerTensor_numpy for details.
* EagerTensor.numpy still returns a copy to ensure no change of behavior.
  This is likely to change in the followup CL.

PiperOrigin-RevId: 246378787

tensorflow/python/BUILD

@@ -443,6 +443,7 @@ cc_library(
         ":numpy_lib",
         ":safe_ptr",
         "//tensorflow/c:c_api",
+        "//tensorflow/c:c_api_internal",
         "//tensorflow/c:tf_status_helper",
         "//tensorflow/core:framework",
         "//tensorflow/core:lib",

tensorflow/python/eager/pywrap_tensor.cc

@@ -627,7 +627,36 @@ static PyObject* EagerTensor_numpy(EagerTensor* self) {
     PyErr_SetString(PyExc_RuntimeError, TF_Message(status.get()));
     return nullptr;
   }
+
+  // HACK(slebedev): The following explains why TensorToNdarray never
+  // reuses the storage.
+  //
+  // TF_TensorToPyArray copies the storage unless its
+  // refcount is 1. For DT_STRING and DT_RESOURCE TF_TensorFromTensor
+  // has to copy so the refcount of the original storage is unchanged.
+  // However, if the storage can be reused by TF_TensorFromTensor its
+  // refcount is +1'd and hence TF_TensorToPyArray no longer can reuse it.
+  //
+  // Here we attempt a direct conversion without an intermediate TF_Tensor
+  // and fall-back to the slow path on failure.
   PyObject* ret = nullptr;
+  if (t->dtype() != tensorflow::DT_STRING &&
+      t->dtype() != tensorflow::DT_RESOURCE) {
+    tensorflow::gtl::InlinedVector<npy_intp, 4> dims(t->dims());
+    for (int d = 0; d < t->dims(); ++d) {
+      dims[d] = t->dim_size(d);
+    }
+
+    auto* copy = new tensorflow::Tensor(*t);
+    char* data = const_cast<char*>(copy->tensor_data().data());
+    if (tensorflow::ArrayFromMemory(
+            dims.size(), dims.data(), data, t->dtype(), [copy] { delete copy; },
+            &ret)
+            .ok()) {
+      return ret;
+    }
+  }
+
   auto cppstatus = tensorflow::TensorToNdarray(*t, &ret);
   if (MaybeRaiseExceptionFromStatus(cppstatus, PyExc_RuntimeError)) {
     Py_XDECREF(ret);

tensorflow/python/eager/tensor_test.py

@@ -487,6 +487,11 @@ class TFETensorUtilTest(test_util.TensorFlowTestCase):
         ValueError, "non-rectangular Python sequence"):
       constant_op.constant(l)
 
+  def test_numpyIsView(self):
+    t = constant_op.constant([0.0])
+    t._numpy()[0] = 42.0
+    self.assertAllClose(t, constant_op.constant([42.0]))
+
 
 if __name__ == "__main__":
   test.main()

tensorflow/python/framework/ops.py

@@ -758,7 +758,8 @@ class _EagerTensorBase(Tensor):
     """
     if self.dtype == dtypes.resource:
       raise ValueError("Resource handles are not convertible to numpy.")
-    return self._cpu_nograd()._numpy()  # pylint: disable=protected-access
+    maybe_arr = self._cpu_nograd()._numpy()  # pylint: disable=protected-access
+    return maybe_arr.copy() if isinstance(maybe_arr, np.ndarray) else maybe_arr
 
   # __int__, __float__ and __index__ may copy the tensor to CPU and
   # only work for scalars; values are cast as per numpy.
@@ -772,7 +773,7 @@ class _EagerTensorBase(Tensor):
     return int(self.numpy())
 
   def __array__(self, dtype=None):
-    return np.array(self.numpy(), dtype=dtype)
+    return np.asarray(self.numpy(), dtype=dtype)
 
   def __format__(self, format_spec):
     return self.numpy().__format__(format_spec)