diff --git a/paddle/fluid/pybind/distributed_py.cc b/paddle/fluid/pybind/distributed_py.cc
index 94b9b36f50adff1598fe3b0ca95ff743f587bb1c..9515ca7f645ec0b5a97b39680cc51d0a3aa36efd 100644
--- a/paddle/fluid/pybind/distributed_py.cc
+++ b/paddle/fluid/pybind/distributed_py.cc
@@ -255,9 +255,9 @@ void BindDistributed(py::module *m) {
                  bool sync_op) {
                 auto out_tensor_list =
                     CastPyArg2VectorOfTensor(py_out_tensor_list.ptr(), 0);
-                Tensor concat_out_tensor = paddle::concat(out_tensor_list, 0);
+                Tensor stack_out_tensor = paddle::stack(out_tensor_list, 0);
                 auto p_out_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_out_tensor.impl());
+                    stack_out_tensor.impl());
                 auto *out_dense = p_out_tensor.get();
 
                 auto in_tensor = CastPyArg2Tensor(py_in_tensor.ptr(), 0);
@@ -307,16 +307,16 @@ void BindDistributed(py::module *m) {
                  bool sync_op) {
                 auto out_tensor_list =
                     CastPyArg2VectorOfTensor(py_out_tensor_list.ptr(), 0);
-                Tensor concat_out_tensor = paddle::concat(out_tensor_list, 0);
+                Tensor stack_out_tensor = paddle::stack(out_tensor_list, 0);
                 auto p_out_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_out_tensor.impl());
+                    stack_out_tensor.impl());
                 auto *out_dense = p_out_tensor.get();
 
                 auto in_tensor_list =
                     CastPyArg2VectorOfTensor(py_in_tensor_list.ptr(), 0);
-                Tensor concat_in_tensor = paddle::concat(in_tensor_list, 0);
+                Tensor stack_in_tensor = paddle::stack(in_tensor_list, 0);
                 auto p_in_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_in_tensor.impl());
+                    stack_in_tensor.impl());
                 auto in_dense = *p_in_tensor;
 
                 // in_tensor_list should not be empty
@@ -430,9 +430,9 @@ void BindDistributed(py::module *m) {
 
                 auto in_tensor_list =
                     CastPyArg2VectorOfTensor(py_in_tensor_list.ptr(), 0);
-                Tensor concat_in_tensor = paddle::concat(in_tensor_list, 0);
+                Tensor stack_in_tensor = paddle::stack(in_tensor_list, 0);
                 auto p_in_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_in_tensor.impl());
+                    stack_in_tensor.impl());
                 auto in_dense = *p_in_tensor;
 
                 distributed::ReduceScatterOptions opts{op};
@@ -484,9 +484,9 @@ void BindDistributed(py::module *m) {
 
                 auto in_tensor_list =
                     CastPyArg2VectorOfTensor(py_in_tensor_list.ptr(), 0);
-                Tensor concat_in_tensor = paddle::concat(in_tensor_list, 0);
+                Tensor stack_in_tensor = paddle::stack(in_tensor_list, 0);
                 auto p_in_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_in_tensor.impl());
+                    stack_in_tensor.impl());
                 auto in_dense = *p_in_tensor;
 
                 distributed::ScatterOptions opts{src};
@@ -746,9 +746,9 @@ void BindDistributed(py::module *m) {
                  py::handle py_in_tensor) {
                 auto out_tensor_list =
                     CastPyArg2VectorOfTensor(py_out_tensor_list.ptr(), 0);
-                Tensor concat_out_tensor = paddle::concat(out_tensor_list, 0);
+                Tensor stack_out_tensor = paddle::stack(out_tensor_list, 0);
                 auto p_out_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_out_tensor.impl());
+                    stack_out_tensor.impl());
                 auto *out_dense = p_out_tensor.get();
 
                 auto in_tensor = CastPyArg2Tensor(py_in_tensor.ptr(), 0);
@@ -854,16 +854,16 @@ void BindDistributed(py::module *m) {
                  py::handle py_in_tensor_list) {
                 auto out_tensor_list =
                     CastPyArg2VectorOfTensor(py_out_tensor_list.ptr(), 0);
-                Tensor concat_out_tensor = paddle::concat(out_tensor_list, 0);
+                Tensor stack_out_tensor = paddle::stack(out_tensor_list, 0);
                 auto p_out_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_out_tensor.impl());
+                    stack_out_tensor.impl());
                 auto *out_dense = p_out_tensor.get();
 
                 auto in_tensor_list =
                     CastPyArg2VectorOfTensor(py_in_tensor_list.ptr(), 0);
-                Tensor concat_in_tensor = paddle::concat(in_tensor_list, 0);
+                Tensor stack_in_tensor = paddle::stack(in_tensor_list, 0);
                 auto p_in_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_in_tensor.impl());
+                    stack_in_tensor.impl());
                 auto in_dense = *p_in_tensor;
 
                 // in_tensor_list should not be empty
@@ -999,9 +999,9 @@ void BindDistributed(py::module *m) {
 
                 auto in_tensor_list =
                     CastPyArg2VectorOfTensor(py_in_tensor_list.ptr(), 0);
-                Tensor concat_in_tensor = paddle::concat(in_tensor_list, 0);
+                Tensor stack_in_tensor = paddle::stack(in_tensor_list, 0);
                 auto p_in_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_in_tensor.impl());
+                    stack_in_tensor.impl());
                 auto in_dense = *p_in_tensor;
 
                 distributed::ReduceScatterOptions opts{op};
@@ -1057,9 +1057,9 @@ void BindDistributed(py::module *m) {
 
                 auto in_tensor_list =
                     CastPyArg2VectorOfTensor(py_in_tensor_list.ptr(), 0);
-                Tensor concat_in_tensor = paddle::concat(in_tensor_list, 0);
+                Tensor stack_in_tensor = paddle::stack(in_tensor_list, 0);
                 auto p_in_tensor = std::dynamic_pointer_cast<phi::DenseTensor>(
-                    concat_in_tensor.impl());
+                    stack_in_tensor.impl());
                 auto in_dense = *p_in_tensor;
 
                 distributed::ScatterOptions opts{src};
diff --git a/paddle/phi/infermeta/multiary.cc b/paddle/phi/infermeta/multiary.cc
index a45a036b298c32e19431a6a0ca8f0f9ea58ed653..319e173adb3b55983f59773e2e5f598ef63eb6f2 100644
--- a/paddle/phi/infermeta/multiary.cc
+++ b/paddle/phi/infermeta/multiary.cc
@@ -911,14 +911,13 @@ void ConcatInferMeta(const std::vector<const MetaTensor*>& x,
   // 1. calculate axis
   int rank = x.at(0)->dims().size();
   PADDLE_ENFORCE_EQ(
-      !rank || (axis >= -rank && axis < rank),
+      axis >= -rank && axis < rank,
       true,
       phi::errors::InvalidArgument(
           "The axis is expected to be in range of [%d, %d), but got %d",
           -rank,
           rank,
           axis));
-  axis = rank ? axis : 0;
   if (axis < 0) {
     axis = axis + rank;
   }
diff --git a/paddle/phi/kernels/cpu/stack_grad_kernel.cc b/paddle/phi/kernels/cpu/stack_grad_kernel.cc
index 018705333e962d6e9852b7937cde58e2aff12cdc..e3190b2c74fbe8afc27f5c4a0f9883214f18795d 100644
--- a/paddle/phi/kernels/cpu/stack_grad_kernel.cc
+++ b/paddle/phi/kernels/cpu/stack_grad_kernel.cc
@@ -54,6 +54,10 @@ PD_REGISTER_KERNEL(stack_grad,
                    phi::StackGradKernel,
                    float,
                    double,
+                   bool,
                    int64_t,
                    int,
+                   uint8_t,
+                   int8_t,
+                   phi::dtype::float16,
                    phi::dtype::bfloat16) {}
diff --git a/paddle/phi/kernels/cpu/stack_kernel.cc b/paddle/phi/kernels/cpu/stack_kernel.cc
index 5eb1cf061be2b072937b8a7da0b882ee4512156c..a9c428c68047d417c09b8a375650b718b4aaa259 100644
--- a/paddle/phi/kernels/cpu/stack_kernel.cc
+++ b/paddle/phi/kernels/cpu/stack_kernel.cc
@@ -57,6 +57,10 @@ PD_REGISTER_KERNEL(stack,
                    phi::StackKernel,
                    float,
                    double,
-                   int,
+                   bool,
                    int64_t,
+                   int,
+                   uint8_t,
+                   int8_t,
+                   phi::dtype::float16,
                    phi::dtype::bfloat16) {}
diff --git a/paddle/phi/kernels/funcs/concat_funcs.h b/paddle/phi/kernels/funcs/concat_funcs.h
index 61a0e6ad7e59ee94710375515650f47a4a75bbd5..db965c2ef9b65413aa8faf2b0dd7813cd8487dd8 100644
--- a/paddle/phi/kernels/funcs/concat_funcs.h
+++ b/paddle/phi/kernels/funcs/concat_funcs.h
@@ -21,14 +21,13 @@ namespace funcs {
 
 static inline int64_t ComputeAxis(int64_t axis, int64_t rank) {
   PADDLE_ENFORCE_EQ(
-      !rank || (axis >= -rank && axis < rank),
+      axis >= -rank && axis < rank,
       true,
       phi::errors::InvalidArgument(
           "The axis is expected to be in range of [%d, %d), but got %d",
           -rank,
           rank,
           axis));
-  axis = rank ? axis : 0;
   if (axis < 0) {
     axis = axis + rank;
   }
diff --git a/paddle/phi/kernels/gpu/concat_kernel.cu b/paddle/phi/kernels/gpu/concat_kernel.cu
index 497f78ca9fc8fa17f4bbf145c4fedb60d8bf5edb..80ff71b2158241370cde44dc581e7b3f388877fa 100644
--- a/paddle/phi/kernels/gpu/concat_kernel.cu
+++ b/paddle/phi/kernels/gpu/concat_kernel.cu
@@ -34,35 +34,6 @@ void ConcatKernel(const Context& dev_ctx,
                   DenseTensor* out) {
   int64_t axis = axis_scalar.to<int64_t>();
 
-  if (UNLIKELY(x[0]->dims().size() == 0)) {
-    // for dims is 0 specially
-    phi::DDim tmp_1dim, out_dims;
-    out_dims[0] = x.size();
-    tmp_1dim[0] = 1;
-
-    out->Resize(out_dims);
-    dev_ctx.template Alloc<T>(out);
-
-    size_t output_offset = 0;
-    for (auto* in : x) {
-      if (in->numel() == 0UL) {
-        continue;
-      }
-      auto in_stride = phi::stride_numel(tmp_1dim);
-      auto out_stride = phi::stride_numel(out->dims());
-      paddle::operators::StridedNumelCopyWithAxis<T>(
-          dev_ctx,
-          axis,
-          out->data<T>() + output_offset,
-          out_stride,
-          in->data<T>(),
-          in_stride,
-          in_stride[axis]);
-      output_offset += in_stride[axis];
-    }
-    return;
-  }
-
   axis = phi::funcs::ComputeAxis(axis, x[0]->dims().size());
 
   std::vector<phi::DDim> x_dims;
diff --git a/paddle/phi/kernels/gpu/stack_grad_kernel.cu b/paddle/phi/kernels/gpu/stack_grad_kernel.cu
index f99747b05933c2b539b25b3693961f959b350932..ea61be0abf1a56ca40c6c52d72b20357f6c2e9e7 100644
--- a/paddle/phi/kernels/gpu/stack_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/stack_grad_kernel.cu
@@ -139,7 +139,10 @@ PD_REGISTER_KERNEL(stack_grad,
                    phi::StackGradKernel,
                    float,
                    double,
+                   bool,
                    int64_t,
                    int,
+                   uint8_t,
+                   int8_t,
                    phi::dtype::float16,
                    phi::dtype::bfloat16) {}
diff --git a/paddle/phi/kernels/gpu/stack_kernel.cu b/paddle/phi/kernels/gpu/stack_kernel.cu
index 5cad80288bf691c3b5c9349d9eb51f1c67f9ab47..3cfb98beca3514dab0707a1556398e67fb1b7d53 100644
--- a/paddle/phi/kernels/gpu/stack_kernel.cu
+++ b/paddle/phi/kernels/gpu/stack_kernel.cu
@@ -175,7 +175,10 @@ PD_REGISTER_KERNEL(stack,
                    phi::StackKernel,
                    float,
                    double,
+                   bool,
                    int64_t,
                    int,
+                   uint8_t,
+                   int8_t,
                    phi::dtype::float16,
                    phi::dtype::bfloat16) {}