Consider corner case.

paddle/operators/expand_op.h

@@ -109,12 +109,24 @@ class ExpandGradKernel : public framework::OpKernel {
     }
 
     int dims = reshape_dims_vec.size() * 6 + reduce_dims_vec.size() - 7;
+    // no need reduce, just copy
+    if (reduce_dims_vec.size() == 0) {
+      auto* in0 = context.Input<Tensor>(framework::GradVarName("Out"));
+      auto* out0 = context.Output<Tensor>(framework::GradVarName("X"));
+      out0->mutable_data<T>(context.GetPlace());
+      if (platform::is_cpu_place(context.GetPlace())) {
+        out0->CopyFrom<T>(*in0, platform::CPUPlace());
+      } else {
+        out0->CopyFrom<T>(*in0, platform::GPUPlace());
+      }
+    } else {
       switch (dims) {
         REP_EXPAND_GRAD_TEMPLATE(72)
         default:
           PADDLE_ENFORCE(false, "Only support tensor whose rank in [1, 6].");
       };
     }
+  }
 
  protected:
   template <int Dims>

python/paddle/v2/framework/tests/test_expand_op.py

@@ -22,8 +22,8 @@ class TestExpandOpRank2(OpTest):
     def setUp(self):
         self.op_type = "expand"
         self.inputs = {'X': np.random.random((12, 14)).astype("float32")}
-        self.attrs = {'expandTimes': [3, 4]}
-        output = np.tile(self.inputs['X'], (3, 4))
+        self.attrs = {'expandTimes': [1, 1]}
+        output = np.tile(self.inputs['X'], (1, 1))
         self.outputs = {'Out': output}
 
     def test_check_output(self):
@@ -37,8 +37,8 @@ class TestExpandOpRank3(OpTest):
     def setUp(self):
         self.op_type = "expand"
         self.inputs = {'X': np.random.random((2, 4, 5)).astype("float32")}
-        self.attrs = {'expandTimes': [3, 2, 1]}
-        output = np.tile(self.inputs['X'], (3, 2, 1))
+        self.attrs = {'expandTimes': [1, 1, 1]}
+        output = np.tile(self.inputs['X'], (1, 1, 1))
         self.outputs = {'Out': output}
 
     def test_check_output(self):