diff --git a/paddle/fluid/operators/concat_op.cc b/paddle/fluid/operators/concat_op.cc
index bbc42d97146f24e69d2f2337967e129af013fb6c..68a52a79e4ce33311780fdf1993397b717a718b2 100644
--- a/paddle/fluid/operators/concat_op.cc
+++ b/paddle/fluid/operators/concat_op.cc
@@ -233,7 +233,8 @@ REGISTER_OP_CPU_KERNEL(
     ops::ConcatKernel<paddle::platform::CPUDeviceContext, int64_t>,
     ops::ConcatKernel<paddle::platform::CPUDeviceContext,
                       paddle::platform::float16>,
-    ops::ConcatKernel<paddle::platform::CPUDeviceContext, int>);
+    ops::ConcatKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ConcatKernel<paddle::platform::CPUDeviceContext, uint8_t>);
 REGISTER_OP_CPU_KERNEL(
     concat_grad,
     ops::ConcatGradKernel<paddle::platform::CPUDeviceContext, double>,
@@ -242,4 +243,5 @@ REGISTER_OP_CPU_KERNEL(
     ops::ConcatGradKernel<paddle::platform::CPUDeviceContext, int64_t>,
     ops::ConcatGradKernel<paddle::platform::CPUDeviceContext,
                           paddle::platform::float16>,
-    ops::ConcatGradKernel<paddle::platform::CPUDeviceContext, int>);
+    ops::ConcatGradKernel<paddle::platform::CPUDeviceContext, int>,
+    ops::ConcatKernel<paddle::platform::CPUDeviceContext, uint8_t>);
diff --git a/paddle/fluid/operators/concat_op.cu.cc b/paddle/fluid/operators/concat_op.cu.cc
index 8c30703f2576b35deb419238de08c5f2fa7b42d2..8732556acb9fdee6e6d83fb34f1bcadf7d8b4bb6 100644
--- a/paddle/fluid/operators/concat_op.cu.cc
+++ b/paddle/fluid/operators/concat_op.cu.cc
@@ -23,7 +23,8 @@ REGISTER_OP_CUDA_KERNEL(
     ops::ConcatKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::ConcatKernel<paddle::platform::CUDADeviceContext, plat::float16>,
     ops::ConcatKernel<paddle::platform::CUDADeviceContext, int64_t>,
-    ops::ConcatKernel<paddle::platform::CUDADeviceContext, int>);
+    ops::ConcatKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ConcatKernel<paddle::platform::CUDADeviceContext, uint8_t>);
 REGISTER_OP_CUDA_KERNEL(
     concat_grad,
     ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, double>,
@@ -31,4 +32,5 @@ REGISTER_OP_CUDA_KERNEL(
     ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, bool>,
     ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, plat::float16>,
     ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, int64_t>,
-    ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, int>);
+    ops::ConcatGradKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::ConcatKernel<paddle::platform::CUDADeviceContext, uint8_t>);
diff --git a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc
index fdb2c57385b2bc1068c618f206bfeb6513d3d8c4..c8d568c8c2cf73041549a138085b72b41c0c297a 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc
+++ b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cc
@@ -100,6 +100,8 @@ REGISTER_OPERATOR(reduce_mean_grad, ops::ReduceGradOp,
                   ops::ReduceMeanDoubleGradOpBaseMaker,
                   ops::ReduceMeanGradNoNeedBufferVarInferer);
 REGISTER_OP_CPU_KERNEL(reduce_mean,
+                       ops::ReduceKernel<paddle::platform::CPUDeviceContext,
+                                         bool, ops::MeanFunctor>,
                        ops::ReduceKernel<paddle::platform::CPUDeviceContext,
                                          float, ops::MeanFunctor>,
                        ops::ReduceKernel<paddle::platform::CPUDeviceContext,
@@ -110,5 +112,6 @@ using CPUReduceMeanGradKernel =
     ops::ReduceGradKernel<paddle::platform::CPUDeviceContext, T,
                           ops::MeanGradFunctor, true>;
 
-REGISTER_OP_CPU_KERNEL(reduce_mean_grad, CPUReduceMeanGradKernel<float>,
+REGISTER_OP_CPU_KERNEL(reduce_mean_grad, CPUReduceMeanGradKernel<bool>,
+                       CPUReduceMeanGradKernel<float>,
                        CPUReduceMeanGradKernel<double>);
diff --git a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu
index cc3653fcb43a4c000d0c61c9d854965fafd59a9c..50d2fcdee23bd9e830f32e0cff4d367c3ad5ba66 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu
+++ b/paddle/fluid/operators/reduce_ops/reduce_mean_op.cu
@@ -65,5 +65,6 @@ class ReduceMeanKernel : public framework::OpKernel<T> {
 }  // namespace operators
 }  // namespace paddle
 
-REGISTER_OP_CUDA_KERNEL(reduce_mean, ops::ReduceMeanKernel<float>,
+REGISTER_OP_CUDA_KERNEL(reduce_mean, ops::ReduceMeanKernel<bool>,
+                        ops::ReduceMeanKernel<float>,
                         ops::ReduceMeanKernel<double>);
diff --git a/paddle/fluid/operators/reduce_ops/reduce_mean_op.part.cu b/paddle/fluid/operators/reduce_ops/reduce_mean_op.part.cu
index 289f574719ff03b1b09f313d05bab152f5c5d651..0e133d5447f93b8891c6de4cb5ad40ac7825493b 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_mean_op.part.cu
+++ b/paddle/fluid/operators/reduce_ops/reduce_mean_op.part.cu
@@ -20,5 +20,6 @@ using CUDAReduceMeanGradKernel =
     ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, T,
                           ops::MeanGradFunctor, true>;
 
-REGISTER_OP_CUDA_KERNEL(reduce_mean_grad, CUDAReduceMeanGradKernel<float>,
+REGISTER_OP_CUDA_KERNEL(reduce_mean_grad, CUDAReduceMeanGradKernel<bool>,
+                        CUDAReduceMeanGradKernel<float>,
                         CUDAReduceMeanGradKernel<double>);
diff --git a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc
index 5a8e8894e1c5da8e0d34f15f2e402b7ecbbea364..a085e851eea77f7bb94076b0f8f83a873b894f3b 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc
+++ b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cc
@@ -109,8 +109,10 @@ REGISTER_OPERATOR(reduce_sum_grad, ops::ReduceGradOp,
                   ops::ReduceSumGradNoNeedBufferVarInferer);
 
 REGISTER_OP_CPU_KERNEL(
-    reduce_sum, ops::ReduceKernel<paddle::platform::CPUDeviceContext, float,
+    reduce_sum, ops::ReduceKernel<paddle::platform::CPUDeviceContext, bool,
                                   ops::SumFunctor>,
+    ops::ReduceKernel<paddle::platform::CPUDeviceContext, float,
+                      ops::SumFunctor>,
     ops::ReduceKernel<paddle::platform::CPUDeviceContext, double,
                       ops::SumFunctor>,
     ops::ReduceKernel<paddle::platform::CPUDeviceContext, int, ops::SumFunctor>,
@@ -128,7 +130,8 @@ using CPUReduceSumGradKernel =
     ops::ReduceSumGradKernel<paddle::platform::CPUDeviceContext, T,
                              ops::SumGradFunctor, true>;
 
-REGISTER_OP_CPU_KERNEL(reduce_sum_grad, CPUReduceSumGradKernel<float>,
+REGISTER_OP_CPU_KERNEL(reduce_sum_grad, CPUReduceSumGradKernel<bool>,
+                       CPUReduceSumGradKernel<float>,
                        CPUReduceSumGradKernel<double>,
                        CPUReduceSumGradKernel<int>,
                        CPUReduceSumGradKernel<int64_t>,
diff --git a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu
index 219cc231a1ea7a0786026d6dcc6d63ce78e24025..dbd020514b2088a336184c4f1ca4f367dd3a14a3 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu
+++ b/paddle/fluid/operators/reduce_ops/reduce_sum_op.cu
@@ -70,7 +70,8 @@ class ReduceSumKernel : public framework::OpKernel<T> {
 }  // namespace operators
 }  // namespace paddle
 
-REGISTER_OP_CUDA_KERNEL(reduce_sum, ops::ReduceSumKernel<float>,
+REGISTER_OP_CUDA_KERNEL(reduce_sum, ops::ReduceSumKernel<bool>,
+                        ops::ReduceSumKernel<float>,
                         ops::ReduceSumKernel<double>, ops::ReduceSumKernel<int>,
                         ops::ReduceSumKernel<int64_t>,
                         ops::ReduceSumKernel<paddle::platform::complex64>,
diff --git a/paddle/fluid/operators/reduce_ops/reduce_sum_op.part.cu b/paddle/fluid/operators/reduce_ops/reduce_sum_op.part.cu
index f2bee6dddc39ec965966e4964c954e5fb1441bf5..67de8bb9a0c1ab4ae917b7e267fc2748087d900e 100644
--- a/paddle/fluid/operators/reduce_ops/reduce_sum_op.part.cu
+++ b/paddle/fluid/operators/reduce_ops/reduce_sum_op.part.cu
@@ -20,7 +20,8 @@ using CUDAReduceSumGradKernel =
     ops::ReduceGradKernel<paddle::platform::CUDADeviceContext, T,
                           ops::SumGradFunctor, true>;
 
-REGISTER_OP_CUDA_KERNEL(reduce_sum_grad, CUDAReduceSumGradKernel<float>,
+REGISTER_OP_CUDA_KERNEL(reduce_sum_grad, CUDAReduceSumGradKernel<bool>,
+                        CUDAReduceSumGradKernel<float>,
                         CUDAReduceSumGradKernel<double>,
                         CUDAReduceSumGradKernel<int>,
                         CUDAReduceSumGradKernel<int64_t>,
diff --git a/python/paddle/tensor/manipulation.py b/python/paddle/tensor/manipulation.py
index 97826f7d5f81d9da8df2c97833f1dcd84a11923a..67e6c7f8e44d740f179961c7e183efdced9ff805 100644
--- a/python/paddle/tensor/manipulation.py
+++ b/python/paddle/tensor/manipulation.py
@@ -80,7 +80,7 @@ def concat(x, axis=0, name=None):
 
     Args:
         x(list|tuple): ``x`` is a Tensor list or Tensor tuple which is with data type bool, float16,
-            float32, float64, int32, int64. All the Tensors in ``x`` must have same data type.
+            float32, float64, int32, int64, uint8. All the Tensors in ``x`` must have same data type.
         axis(int|Tensor, optional): Specify the axis to operate on the input Tensors.
             It's a scalar with data type int or a Tensor with shape [1] and data type int32 
             or int64. The effective range is [-R, R), where R is Rank(x). When ``axis < 0``,