diff --git a/paddle/fluid/operators/activation_op.cc b/paddle/fluid/operators/activation_op.cc
index c53427b465bc3cefe2eb9d539433eef13c0eee74..348902c656cec1ea1eeaccc90feefd56d307111d 100644
--- a/paddle/fluid/operators/activation_op.cc
+++ b/paddle/fluid/operators/activation_op.cc
@@ -602,7 +602,6 @@ REGISTER_ACTIVATION_OP_MAKER(Softsign, SoftsignDoc);
 
 namespace ops = paddle::operators;
 
-
 #define REGISTER_ACTIVATION_OP(KERNEL_TYPE, OP_NAME, functor, grad_functor) \
   REGISTER_OPERATOR(                                                        \
       KERNEL_TYPE, ops::ActivationOp, ops::OP_NAME##OpMaker,                \
@@ -619,7 +618,6 @@ namespace ops = paddle::operators;
 
 #define REGISTER_ACTIVATION_CPU_KERNEL(act_type, op_name, functor,        \
                                        grad_functor)                      \
-
   REGISTER_OP_CPU_KERNEL(                                                 \
       act_type, ops::ActivationKernel<paddle::platform::CPUDeviceContext, \
                                       ops::functor<float>>,               \
diff --git a/paddle/fluid/operators/activation_op.h b/paddle/fluid/operators/activation_op.h
index e50f3bf766d139a43b95fdae2b9e48e8761cc87a..1732f61582f79365d6872e15b9df1ee8f053903c 100644
--- a/paddle/fluid/operators/activation_op.h
+++ b/paddle/fluid/operators/activation_op.h
@@ -36,7 +36,6 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
-
 enum ActBwdOpFwdDeps {
   kNoDeps = 0x00,  // Do not need any forward input/output
   kDepX = 0x01,    // Only need forward input X
@@ -528,6 +527,8 @@ struct RsqrtGradFunctor : public BaseActivationFunctor<T> {
   void operator()(Device d, X x, Out out, dOut dout, dX dx) const {
     dx.device(d) = static_cast<T>(-0.5) * dout * out * out * out;
   }
+
+  static constexpr ActBwdOpFwdDeps FwdDeps() { return kDepOut; }
 };
 
 // ceil(x) = ceiling(x)
@@ -1200,7 +1201,6 @@ struct SwishGradFunctor : public BaseActivationFunctor<T> {
 }  // namespace operators
 }  // namespace paddle
 
-
 #define FOR_EACH_ACTIVATION_OP(__macro)                                       \
   __macro(sigmoid, Sigmoid, SigmoidFunctor, SigmoidGradFunctor);              \
   __macro(logsigmoid, LogSigmoid, LogSigmoidFunctor, LogSigmoidGradFunctor);  \
@@ -1211,6 +1211,7 @@ struct SwishGradFunctor : public BaseActivationFunctor<T> {
   __macro(atan, Atan, AtanFunctor, AtanGradFunctor);                          \
   __macro(softshrink, SoftShrink, SoftShrinkFunctor, SoftShrinkGradFunctor);  \
   __macro(sqrt, Sqrt, SqrtFunctor, SqrtGradFunctor);                          \
+  __macro(rsqrt, Rsqrt, RsqrtFunctor, RsqrtGradFunctor);                      \
   __macro(abs, Abs, AbsFunctor, AbsGradFunctor);                              \
   __macro(ceil, Ceil, CeilFunctor, ZeroGradFunctor);                          \
   __macro(floor, Floor, FloorFunctor, ZeroGradFunctor);                       \