diff --git a/lite/kernels/x86/elementwise_compute.cc b/lite/kernels/x86/elementwise_compute.cc
index 8b375f078f6..01a830963cc 100644
--- a/lite/kernels/x86/elementwise_compute.cc
+++ b/lite/kernels/x86/elementwise_compute.cc
@@ -47,6 +47,28 @@ REGISTER_LITE_KERNEL(elementwise_mul,
     .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86))})
     .Finalize();
 
+REGISTER_LITE_KERNEL(elementwise_mul,
+                     kX86,
+                     kFloat,
+                     kNCHW,
+                     paddle::lite::kernels::x86::ElementwiseMulCompute<int>,
+                     int32)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt32))})
+    .BindInput("Y", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt32))})
+    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt32))})
+    .Finalize();
+
+REGISTER_LITE_KERNEL(elementwise_mul,
+                     kX86,
+                     kFloat,
+                     kNCHW,
+                     paddle::lite::kernels::x86::ElementwiseMulCompute<int64_t>,
+                     int64)
+    .BindInput("X", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt64))})
+    .BindInput("Y", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt64))})
+    .BindOutput("Out", {LiteType::GetTensorTy(TARGET(kX86), PRECISION(kInt64))})
+    .Finalize();
+
 REGISTER_LITE_KERNEL(
     elementwise_floordiv,
     kX86,
diff --git a/lite/tests/kernels/elementwise_compute_test.cc b/lite/tests/kernels/elementwise_compute_test.cc
index d583c95f15b..460a39f2331 100644
--- a/lite/tests/kernels/elementwise_compute_test.cc
+++ b/lite/tests/kernels/elementwise_compute_test.cc
@@ -372,6 +372,9 @@ TEST(elementwise_x86, precison) {
   Place place(TARGET(kX86));
   float abs_error = 1e-5;
 
+  TestEltX86<float>(place, abs_error, "mul", "def");
+  TestEltX86<int>(place, abs_error, "mul", "int32");
+  TestEltX86<int64_t>(place, abs_error, "mul", "int64");
   TestEltX86<float>(place, abs_error, "floordiv", "float32");
   TestEltX86<int32_t>(place, abs_error, "floordiv", "int32");
   TestEltX86<int64_t>(place, abs_error, "floordiv", "int64");
diff --git a/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc b/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc
index e7196fc04c0..1fd47744947 100644
--- a/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc
+++ b/lite/tests/kernels/fill_constant_batch_size_like_compute_test.cc
@@ -132,7 +132,9 @@ TEST(fill_constant_batch_size_like, precision) {
   LOG(INFO) << "test fill_constant_batch_size_like op";
   Place place;
   float abs_error = 1e-5;
-#if defined(LITE_WITH_NPU)
+#if defined(LITE_WITH_XPU) && !defined(LITE_WITH_XTCL)
+  place = TARGET(kXPU);
+#elif defined(LITE_WITH_NPU)
   place = TARGET(kNPU);
   abs_error = 1e-2;  // use fp16 in npu
 #elif defined(LITE_WITH_ARM) || defined(LITE_WITH_X86)