diff --git a/paddle/phi/backends/xpu/xpu2_op_list.cc b/paddle/phi/backends/xpu/xpu2_op_list.cc
index aa9714ee36eaaf296c22587584181140d4a49c68..3a7a0f2fd6b5636f10f6d72be58ddcd29e850b4c 100644
--- a/paddle/phi/backends/xpu/xpu2_op_list.cc
+++ b/paddle/phi/backends/xpu/xpu2_op_list.cc
@@ -603,7 +603,17 @@ XPUOpMap& get_kl2_ops() {
                      phi::DataType::BOOL,
                      phi::DataType::INT8,
                      phi::DataType::UINT8,
+                     phi::DataType::FLOAT16,
                      phi::DataType::FLOAT32})},
+      {"unsqueeze_with_xshape",
+       XPUKernelSet({phi::DataType::FLOAT64,
+                     phi::DataType::INT64,
+                     phi::DataType::INT32,
+                     phi::DataType::BOOL,
+                     phi::DataType::INT8,
+                     phi::DataType::UINT8,
+                     phi::DataType::FLOAT32,
+                     phi::DataType::FLOAT16})},
       {"warpctc_grad", XPUKernelSet({phi::DataType::FLOAT32})},
       {"warpctc", XPUKernelSet({phi::DataType::FLOAT32})},
       {"where_index",
diff --git a/paddle/phi/kernels/xpu/cross_entropy_kernel.cc b/paddle/phi/kernels/xpu/cross_entropy_kernel.cc
index f054c6c445148c47dc459761dc6ce7f6c5aef247..f1b2257427f67b1ce8bc3dc194a4d6a3975246a4 100644
--- a/paddle/phi/kernels/xpu/cross_entropy_kernel.cc
+++ b/paddle/phi/kernels/xpu/cross_entropy_kernel.cc
@@ -89,10 +89,15 @@ void CrossEntropyWithSoftmaxKernel(const Context& dev_ctx,
                             static_cast<XPUType>(max_val));
   PADDLE_ENFORCE_XDNN_SUCCESS(r, "clip_v2");
 
-  r = xpu::softmax<XPUType>(
-      dev_ctx.x_context(), clip_logits_data, softmax_data, logits_dims, axis);
-  PADDLE_ENFORCE_XDNN_SUCCESS(r, "softmax");
-
+  if (use_softmax) {
+    r = xpu::softmax<XPUType>(
+        dev_ctx.x_context(), clip_logits_data, softmax_data, logits_dims, axis);
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "softmax");
+  } else {
+    r = xpu::copy<XPUType>(
+        dev_ctx.x_context(), clip_logits_data, softmax_data, softmax->numel());
+    PADDLE_ENFORCE_XDNN_SUCCESS(r, "copy");
+  }
   // cross_entropy
   if (axis != rank - 1) {
     XPUType* trans_softmax = RAII_GUARD.alloc_l3_or_gm<XPUType>(n * d);