fix the dynamic shape with heterogeneity

mindspore/ccsrc/runtime/graph_scheduler/actor/custom_actor.cc

@@ -39,9 +39,9 @@ void CustomActor::Run(OpContext<DeviceTensor> *const ctx) {
     std::string error_info = "Launch custom kernel exception: " + node->fullname_with_scope();
     SET_OPCONTEXT_FAIL_RET_WITH_ERROR_BY_STRATEGY(strategy_, (*ctx), error_info);
   }
-  // update the output addr size after initop && updateop, because after the initop & updateop, the shape of output
+  // Update the output addr size after inferop && updateop, because after the inferop & updateop, the shape of output
-  // maybe changed
+  // maybe changed.
-  if (AnfUtils::GetCustomActorType(kernel_.lock()) == kInit ||
+  if (AnfUtils::GetCustomActorType(kernel_.lock()) == kInfer ||
       AnfUtils::GetCustomActorType(kernel_.lock()) == kUpdate) {
     auto base_node = AnfUtils::GetCustomActorBaseNode(kernel_.lock());
     auto kernel_info = dynamic_cast<KernelInfo *>(base_node->kernel_info());

tests/st/dynamic_shape/test_dynamic_shape_with_heterogeneity.py

@@ -44,6 +44,19 @@ class UniqueSquare(nn.Cell):
         return self.square(x)
 
 
+class UniqueSquareRelu(nn.Cell):
+    def __init__(self):
+        super(UniqueSquareRelu, self).__init__()
+        self.unique_cpu = P.Unique().add_prim_attr("primitive_target", "CPU")
+        self.square_cpu = P.Square().add_prim_attr("primitive_target", "CPU")
+        self.relu = P.ReLU()
+
+    def construct(self, x):
+        x, _ = self.unique_cpu(x)
+        x = self.square_cpu(x)
+        return self.relu(x)
+
+
 class UniqueReshapeAdd(nn.Cell):
     def __init__(self):
         super(UniqueReshapeAdd, self).__init__()
@@ -99,6 +112,24 @@ def test_unique_square():
     assert (output.asnumpy() == expect).all()
 
 
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.platform_arm_ascend_training
+@pytest.mark.platform_x86_ascend_training
+@pytest.mark.env_onecard
+def test_unique_square_relu():
+    """
+    Feature: Dynamic shape with heterogeneity.
+    Description: Test unique, square and relu kernels in dynamic shape with heterogeneity scenarios.
+    Expectation: The value and shape of output are the expected values.
+    """
+    x = Tensor(np.array([1, 1, 2, 2, 3, 3]), mstype.float32)
+    net = UniqueSquareRelu()
+    output = net(x)
+    expect = np.array([1, 4, 9])
+    assert (output.asnumpy() == expect).all()
+
+
 @pytest.mark.level0
 @pytest.mark.platform_x86_gpu_training
 @pytest.mark.platform_arm_ascend_training