diff --git a/paddle/phi/kernels/xpu/concat_and_split_functor.cc b/paddle/phi/kernels/xpu/concat_and_split_functor.cc
index edff1e90143b63796825364c79c0636d30170786..a1335f33b67007c0182869eb2f463fb673511a5b 100644
--- a/paddle/phi/kernels/xpu/concat_and_split_functor.cc
+++ b/paddle/phi/kernels/xpu/concat_and_split_functor.cc
@@ -85,12 +85,21 @@ class SplitFunctor<XPUContext, T> {
 
     int num = ins.size();
     auto input_dims = ins[0]->dims();
+    // special for 0-dim shape
+    if (input_dims.size() == 0) {
+      input_dims = {1};
+    }
     std::vector<int> split_list(num);
     std::vector<int> xdims_list(input_dims.size());
     int total_length = 0;
     for (int i = 0; i < num; ++i) {
-      split_list[i] = ins[i]->dims()[axis];
-      total_length += ins[i]->dims()[axis];
+      auto ins_i_dims = ins[i]->dims();
+      // special for 0-dim shape
+      if (ins_i_dims.size() == 0) {
+        ins_i_dims = {1};
+      }
+      split_list[i] = ins_i_dims[axis];
+      total_length += ins_i_dims[axis];
     }
 
     for (int i = 0; i < input_dims.size(); ++i) {
@@ -110,6 +119,9 @@ class SplitFunctor<XPUContext, T> {
       context.template Alloc<T>(&tmp_data);
     }
 
+    // int split(Context* ctx, const T* x, const std::vector<T*>& y_list, const
+    // std::vector<int64_t>& xshape, const std::vector<int64_t>& split_list,
+    // int64_t axis);
     auto r = xpu::split<XPUType>(
         context.x_context(),
         reinterpret_cast<const XPUType*>(tmp_data.data<T>()),