diff --git a/paddle/fluid/operators/abs_op.cu b/paddle/fluid/operators/abs_op.cu
index 94b0a3ae729380c90f1351df3b2d42d34d88be57..86748d4505d28708ed2d803edc1d52501889e51e 100644
--- a/paddle/fluid/operators/abs_op.cu
+++ b/paddle/fluid/operators/abs_op.cu
@@ -24,14 +24,14 @@ struct CudaAbsFunctor;
 
 template <typename T>
 struct CudaAbsFunctor<T, math::Complex<T, math::Real<T>>> {
-  __device__ __forceinline__ math::Real<T> operator()(const T& x) const {
+  __device__ __forceinline__ math::Real<T> operator()(const T x) const {
     return abs(x);
   }
 };
 
 template <typename T>
 struct CudaAbsFunctor<T, math::NoComplex<T, math::Real<T>>> {
-  __device__ __forceinline__ T operator()(const T& x) const {
+  __device__ __forceinline__ T operator()(const T x) const {
     return std::abs(x);
   }
 };
diff --git a/paddle/fluid/operators/bce_loss_op.cu b/paddle/fluid/operators/bce_loss_op.cu
index 18562b243255be9077cb2310de3d7f1d4857e969..da96aa92cd25a96d250ed5c2b86abf42dd16b927 100644
--- a/paddle/fluid/operators/bce_loss_op.cu
+++ b/paddle/fluid/operators/bce_loss_op.cu
@@ -28,8 +28,8 @@ template <typename T>
 struct BCELossGradFunctor {
   T one = static_cast<T>(1.0f);
   T eps = static_cast<T>(1e-12);
-  __device__ __forceinline__ T operator()(const T& x, const T& label,
-                                          const T& dout) const {
+  __device__ __forceinline__ T operator()(const T x, const T label,
+                                          const T dout) const {
     T term1 = max((one - x) * x, eps);
     return (dout * (x - label) / term1);
   }
diff --git a/paddle/fluid/operators/clip_op.h b/paddle/fluid/operators/clip_op.h
index f08a7b2d573145ecc057f70aa5c8c02465746da8..3672fa983e495cb6a4ec5b5fb488a7a5a40fb9b2 100644
--- a/paddle/fluid/operators/clip_op.h
+++ b/paddle/fluid/operators/clip_op.h
@@ -32,7 +32,7 @@ template <typename T>
 class ClipFunctor {
  public:
   explicit ClipFunctor(const T min, const T max) : min_(min), max_(max) {}
-  HOSTDEVICE T operator()(const T& x) const {
+  HOSTDEVICE T operator()(const T x) const {
     return x < min_ ? min_ : x > max_ ? max_ : x;
   }
 
diff --git a/paddle/fluid/operators/p_norm_op.cu b/paddle/fluid/operators/p_norm_op.cu
index 1db6f6e51746282f89e9c160a0d370b72eecd605..b2a9ca6f93742730173b9268abc4ede301d755d4 100644
--- a/paddle/fluid/operators/p_norm_op.cu
+++ b/paddle/fluid/operators/p_norm_op.cu
@@ -63,7 +63,7 @@ __device__ __forceinline__ double inline_pow(double base, double exponent) {
 template <typename T>
 struct NonzeroFunctor {
   HOSTDEVICE explicit inline NonzeroFunctor() {}
-  HOSTDEVICE inline T operator()(const T& x) const {
+  HOSTDEVICE inline T operator()(const T x) const {
     return static_cast<T>(static_cast<double>(x) != 0);
   }
 };
@@ -71,7 +71,7 @@ struct NonzeroFunctor {
 template <typename T>
 struct AbsFunctor {
   HOSTDEVICE explicit inline AbsFunctor() {}
-  HOSTDEVICE inline T operator()(const T& x) const {
+  HOSTDEVICE inline T operator()(const T x) const {
     return static_cast<T>(inline_abs(x));
   }
 };
@@ -81,7 +81,7 @@ struct UnsignedPowFunctor {
   HOSTDEVICE explicit inline UnsignedPowFunctor(float porder) {
     this->porder = porder;
   }
-  HOSTDEVICE inline Ty operator()(const Tx& x) const {
+  HOSTDEVICE inline Ty operator()(const Tx x) const {
     return static_cast<Ty>(inline_pow(inline_abs(x), static_cast<Tx>(porder)));
   }
   float porder;
@@ -90,7 +90,7 @@ struct UnsignedPowFunctor {
 template <typename Tx, typename Ty = Tx>
 struct PowFunctor {
   HOSTDEVICE explicit inline PowFunctor(float porder) { this->porder = porder; }
-  HOSTDEVICE inline Ty operator()(const Tx& x) const {
+  HOSTDEVICE inline Ty operator()(const Tx x) const {
     return static_cast<Ty>(inline_pow(x, static_cast<Tx>(porder)));
   }
   float porder;
diff --git a/paddle/fluid/operators/renorm_op.cu b/paddle/fluid/operators/renorm_op.cu
index 1798faa759bed013c35f93bf0d2623f2a17da49a..b21b9fde56f247b5a26f1b9b600fe5da76111627 100644
--- a/paddle/fluid/operators/renorm_op.cu
+++ b/paddle/fluid/operators/renorm_op.cu
@@ -42,7 +42,7 @@ struct UnsignedPowFunctor {
   HOSTDEVICE explicit inline UnsignedPowFunctor(float porder) {
     this->porder = porder;
   }
-  HOSTDEVICE inline Ty operator()(const Tx& x) const {
+  HOSTDEVICE inline Ty operator()(const Tx x) const {
     return static_cast<Ty>(inline_pow(inline_abs(x), static_cast<Tx>(porder)));
   }
   float porder;
diff --git a/paddle/pten/kernels/gpu/cast_kernel.cu b/paddle/pten/kernels/gpu/cast_kernel.cu
index 9f65400f93b9f0cbac0e8aae41fa4678b52a8bfa..0bbe7a3a132d1c0737d174a9a0c99b22fd48a36a 100644
--- a/paddle/pten/kernels/gpu/cast_kernel.cu
+++ b/paddle/pten/kernels/gpu/cast_kernel.cu
@@ -30,7 +30,7 @@ namespace pten {
 
 template <typename InT, typename OutT>
 struct CastFuctor {
-  __device__ __forceinline__ OutT operator()(const InT& x) const {
+  __device__ __forceinline__ OutT operator()(const InT x) const {
     return static_cast<OutT>(x);
   }
 };
diff --git a/paddle/pten/kernels/gpu/scale_kernel.cu b/paddle/pten/kernels/gpu/scale_kernel.cu
index f4bb5c5dbf75502bfad09c987d3da22864d99403..68574c063e77f34988e468c5f46fcbfabba22b88 100644
--- a/paddle/pten/kernels/gpu/scale_kernel.cu
+++ b/paddle/pten/kernels/gpu/scale_kernel.cu
@@ -34,7 +34,7 @@ struct ScaleFunctor {
     bias_after_scale = is_bias_after_sacle;
   }
 
-  __device__ __forceinline__ InT operator()(const InT& x) const {
+  __device__ __forceinline__ InT operator()(const InT x) const {
     if (bias_after_scale) {
       return scale * x + bias;
     } else {