set data permits requires_grad=True on integer tensor

Pull Request resolved: https://github.com/pytorch/pytorch/pull/78436

Approved by: https://github.com/albanD, https://github.com/soulitzer

test/custom_operator/test_custom_ops.cpp

@@ -164,10 +164,10 @@ void test_move_to_dtype(const std::string& path_to_exported_script_module) {
   torch::jit::Module module =
       torch::jit::load(path_to_exported_script_module);
 
-  module.to(torch::kInt);
+  module.to(torch::kFloat16);
 
   helpers::check_all_parameters(module, [](const torch::Tensor& tensor) {
-    return tensor.dtype() == torch::kInt;
+    return tensor.dtype() == torch::kFloat16;
   });
 
   module.to(torch::kDouble);

test/test_autograd.py

@@ -4301,6 +4301,14 @@ class TestAutograd(TestCase):
         b.data = a
         self.assertTrue(b_id_saved == id(b))
 
+    def test_set_data_self_requires_grad(self):
+        a = torch.tensor(1.0, requires_grad=True)
+        b = torch.tensor(2.0)
+        c = torch.tensor(3, dtype=torch.int64)
+        a.data = b
+        with self.assertRaisesRegex(RuntimeError, 'must be floating point or complex dtype'):
+            a.data = c
+
     @unittest.skipIf(IS_WINDOWS, "Skipping because doesn't work for windows")
     def test_thread_shutdown(self):
         code = """import torch

test/test_torch.py

@@ -7535,12 +7535,12 @@ tensor([[[1.+1.j, 1.+1.j, 1.+1.j,  ..., 1.+1.j, 1.+1.j, 1.+1.j],
         # raise a nice error.
         with self.assertRaisesRegex(
                 RuntimeError,
-                # message includes both Double and Long
-                '(?=.*Double)(?=.*Long)'):
+                # message includes both Double and ComplexFloat
+                '(?=.*Double)(?=.*ComplexFloat)'):
 
             # Calls model with a LongTensor input but DoubleTensor weights
             input = torch.randn(1, 1, 1, 6, dtype=torch.double)
-            weight = torch.zeros(1, 1, 1, 3, dtype=torch.long)
+            weight = torch.zeros(1, 1, 1, 3, dtype=torch.complex64)
             model = torch.nn.Conv2d(1, 1, (1, 3), stride=1, padding=0, bias=False)
             model.weight.data = weight
             out = model(input)

torch/csrc/autograd/variable.cpp

@@ -401,6 +401,10 @@ void VariableHooks::set_data(const at::TensorBase & self_base, const at::TensorB
     _has_compatible_shallow_copy_type(self, new_data),
     "Attempted to call `variable.set_data(tensor)`, but `variable` and `tensor` have incompatible tensor type.");
 
+  TORCH_CHECK(
+    !self.requires_grad() || isDifferentiableType(at::typeMetaToScalarType(new_data.dtype())),
+    "data set to a tensor that requires gradients must be floating point or complex dtype");
+
   // Resets gradient accumulator if metadata is out of date
   AutogradMeta* autograd_meta = impl::get_autograd_meta(self);
   if (autograd_meta) {