Numpy zerotensor handling (#164487)

Fixes #89034

Updated tensor_to_numpy() function in tensor_numpy.cpp to handle ZeroTensors by throwing an error if force=False and returning an array full of zeros if force=True.

@ngimel, I just saw that you mentioned PyTorch is not too concerned with this issue but I had already worked on it so I figured I would push it anyways and see what you thought. Feel free to close the PR if you think it is not worth merging.

@albanD

Pull Request resolved: https://github.com/pytorch/pytorch/pull/164487
Approved by: https://github.com/ngimel, https://github.com/albanD

test/test_numpy_interop.py

@@ -164,6 +164,28 @@ class TestNumPyInterop(TestCase):
         self.assertEqual(y.dtype, np.bool_)
         self.assertEqual(x[0], y[0])
 
+    @skipIfTorchDynamo(
+        "can't check if value is ZeroTensor since _is_zerotensor returns a bool and not a TensorVariable"
+    )
+    def test_to_numpy_zero_tensor(self, device) -> None:
+        dtypes = [
+            torch.uint8,
+            torch.int8,
+            torch.short,
+            torch.int,
+            torch.half,
+            torch.float,
+            torch.double,
+            torch.long,
+            torch.bool,
+        ]
+        for dtype in dtypes:
+            x = torch._efficientzerotensor((10), dtype=dtype)
+            self.assertRaises(RuntimeError, lambda: x.numpy())
+            y = x.numpy(force=True)
+            for i in range(10):
+                self.assertEqual(y[i], 0)
+
     @skipIfTorchDynamo("conj bit not implemented in TensorVariable yet")
     def test_to_numpy_force_argument(self, device) -> None:
         for force in [False, True]:

torch/csrc/utils/tensor_numpy.cpp

@@ -145,6 +145,10 @@ PyObject* tensor_to_numpy(const at::Tensor& tensor, bool force /*=false*/) {
         " device type tensor to numpy. Use Tensor.cpu() to ",
         "copy the tensor to host memory first.");
 
+    TORCH_CHECK(
+        !at::_is_zerotensor(tensor),
+        " Cannot convert a ZeroTensor to numpy. Set force=True if you need the zero array.");
+
     TORCH_CHECK(
         !(at::GradMode::is_enabled() && tensor.requires_grad()),
         "Can't call numpy() on Tensor that requires grad. "
@@ -186,6 +190,9 @@ PyObject* tensor_to_numpy(const at::Tensor& tensor, bool force /*=false*/) {
   if (!array)
     return nullptr;
 
+  if (at::_is_zerotensor(tensor))
+    PyArray_FILLWBYTE(reinterpret_cast<PyArrayObject*>(array.get()), 0);
+
   // TODO: This attempts to keep the underlying memory alive by setting the base
   // object of the ndarray to the tensor and disabling resizes on the storage.
   // This is not sufficient. For example, the tensor's storage may be changed