diff --git a/paddle/operators/math/maxouting.cc b/paddle/operators/math/maxouting.cc
index c8c1974f79a94563e930caebe63790bb1d5adc17..bcd4da612c841e241b63e62ce19e7c5f4a7db2d5 100644
--- a/paddle/operators/math/maxouting.cc
+++ b/paddle/operators/math/maxouting.cc
@@ -89,6 +89,7 @@ public:
               if (input_data[input_idx] == output_data[output_idx]) {
                 input_grad_data[input_idx] += output_grad_data[output_idx];
                 continue_match = false;
+                break;
               }
           }
         }
diff --git a/paddle/operators/math/maxouting.cu b/paddle/operators/math/maxouting.cu
index 3a0600fd8465e5183ded3775afb459f27e3da0bf..0a8afbbaca3a793b62bc37fd5997a0f6213661f5 100644
--- a/paddle/operators/math/maxouting.cu
+++ b/paddle/operators/math/maxouting.cu
@@ -65,6 +65,7 @@ __global__ void KernelMaxoutGrad(
         if (input_data[data_idx + g * feat_len] == output_data[i]) {
           max_index = data_idx + g * feat_len;
           continue_match = false;
+          break;
         }
       }
       if (max_index != -1) {
diff --git a/paddle/operators/maxout_op.cu.cc b/paddle/operators/maxout_op.cu.cc
index 3e6debf69960be1a7ebd2c00505b31ec7484c40a..5ee431cb26cde12a620808fb07dbd26843c64bc5 100644
--- a/paddle/operators/maxout_op.cu.cc
+++ b/paddle/operators/maxout_op.cu.cc
@@ -17,6 +17,11 @@
 namespace ops = paddle::operators;
 REGISTER_OP_GPU_KERNEL(maxout, ops::MaxOutKernel<paddle::platform::GPUPlace,
                        float>);
+REGISTER_OP_GPU_KERNEL(maxout, ops::MaxOutKernel<paddle::platform::GPUPlace,
+                       double>);
 REGISTER_OP_GPU_KERNEL(maxout_grad,
                        ops::MaxOutGradKernel<paddle::platform::GPUPlace,
                        float>);
+REGISTER_OP_GPU_KERNEL(maxout_grad,
+                       ops::MaxOutGradKernel<paddle::platform::GPUPlace,
+                       double>);