diff --git a/paddle/phi/kernels/xpu/index_select_kernel.cc b/paddle/phi/kernels/xpu/index_select_kernel.cc
index e3eb974fe172e..75c19aa028bce 100644
--- a/paddle/phi/kernels/xpu/index_select_kernel.cc
+++ b/paddle/phi/kernels/xpu/index_select_kernel.cc
@@ -40,29 +40,33 @@ void IndexSelectKernel(const Context& ctx,
                         index_type,
                         phi::DataType::INT32,
                         phi::DataType::INT64));
-
   auto* in_data = x.data<T>();
   std::vector<int> in_shape = phi::vectorize<int>(input_dim);
   int index_len = output->dims()[dim];
   T* out_data = ctx.template Alloc<T>(output);
   int r = 0;
   xpu::ctx_guard RAII_GUARD(ctx.x_context());
-  const int8_t* index_ptr = nullptr;
-  int byte_times = sizeof(index_type);
+  int8_t* index_ptr = nullptr;  // temp xpu buffer
+  int byte_times = SizeOf(index_type);
   if (index.place() == CPUPlace()) {
     index_ptr = RAII_GUARD.alloc_l3_or_gm<int8_t>(byte_times * index.numel());
     PADDLE_ENFORCE_XDNN_NOT_NULL(index_ptr);
+    const void* cpu_idx_data = nullptr;
+    if (index_type == phi::DataType::INT64) {
+      cpu_idx_data = reinterpret_cast<const void*>(index.data<int64_t>());
+    } else if (index_type == phi::DataType::INT32) {
+      cpu_idx_data = reinterpret_cast<const void*>(index.data<int>());
+    }
     memory_utils::Copy(ctx.GetPlace(),
-                       reinterpret_cast<void*>(const_cast<int8_t*>(index_ptr)),
+                       reinterpret_cast<void*>(index_ptr),
                        CPUPlace(),
-                       reinterpret_cast<const void*>(index.data<int>()),
+                       cpu_idx_data,
                        byte_times * index.numel());
-  } else {
-    index_ptr = index.template data<int8_t>();
   }
   if (index_type == phi::DataType::INT64) {
     const int64_t* index_data =
-        reinterpret_cast<const int64_t*>(const_cast<int8_t*>(index_ptr));
+        index_ptr ? reinterpret_cast<const int64_t*>(index_ptr)
+                  : index.template data<int64_t>();
     r = xpu::gather<T, int64_t>(ctx.x_context(),
                                 in_data,
                                 index_data,
@@ -71,8 +75,8 @@ void IndexSelectKernel(const Context& ctx,
                                 index_len,
                                 dim);
   } else {
-    const int* index_data =
-        reinterpret_cast<const int*>(const_cast<int8_t*>(index_ptr));
+    const int* index_data = index_ptr ? reinterpret_cast<const int*>(index_ptr)
+                                      : index.template data<int>();
     r = xpu::gather<T, int>(ctx.x_context(),
                             in_data,
                             index_data,