diff --git a/paddle/fluid/operators/batch_norm_op.cu.cc b/paddle/fluid/operators/batch_norm_op.cu.cc
index 2d1556efc66826ea9847de8311ccecdee0ea7871..949497f48c666930c97cc2e16c12164408b25bdc 100644
--- a/paddle/fluid/operators/batch_norm_op.cu.cc
+++ b/paddle/fluid/operators/batch_norm_op.cu.cc
@@ -270,9 +270,9 @@ class BatchNormGradKernel<platform::CUDADeviceContext, T>
 }  // namespace paddle
 
 namespace ops = paddle::operators;
+namespace plat = paddle::platform;
 REGISTER_OP_CUDA_KERNEL(
-    batch_norm,
-    ops::BatchNormKernel<paddle::platform::CUDADeviceContext, float>);
+    batch_norm, ops::BatchNormKernel<plat::CUDADeviceContext, float>,
+    ops::BatchNormKernel<plat::CUDADeviceContext, plat::float16>);
 REGISTER_OP_CUDA_KERNEL(
-    batch_norm_grad,
-    ops::BatchNormGradKernel<paddle::platform::CUDADeviceContext, float>);
+    batch_norm_grad, ops::BatchNormGradKernel<plat::CUDADeviceContext, float>);
diff --git a/paddle/fluid/operators/math/math_function.cc b/paddle/fluid/operators/math/math_function.cc
index 35d251f71a0cb631d5900498ea3188b5ddeae334..1cbd2fa870165bf9416c44d69ed66170a1737009 100644
--- a/paddle/fluid/operators/math/math_function.cc
+++ b/paddle/fluid/operators/math/math_function.cc
@@ -278,6 +278,7 @@ void axpy<platform::CPUDeviceContext, double>(
   cblas_daxpy(n, alpha, x, 1, y, 1);
 }
 
+template struct SetConstant<platform::CPUDeviceContext, platform::float16>;
 template struct SetConstant<platform::CPUDeviceContext, float>;
 template struct SetConstant<platform::CPUDeviceContext, double>;
 template struct SetConstant<platform::CPUDeviceContext, int>;
diff --git a/paddle/fluid/operators/math/math_function.cu b/paddle/fluid/operators/math/math_function.cu
index 3abbcdb71d03eaf6f8eba3d97150d27ac5a5405e..bccfaef9ce22d05aa00d81c3933987f6ab66b5c8 100644
--- a/paddle/fluid/operators/math/math_function.cu
+++ b/paddle/fluid/operators/math/math_function.cu
@@ -348,6 +348,7 @@ void axpy<platform::CUDADeviceContext, double>(
                                                 &alpha, x, 1, y, 1));
 }
 
+template struct SetConstant<platform::CUDADeviceContext, platform::float16>;
 template struct SetConstant<platform::CUDADeviceContext, float>;
 template struct SetConstant<platform::CUDADeviceContext, double>;
 template struct SetConstant<platform::CUDADeviceContext, int>;