diff --git a/src/operators/kernel/central-arm-func/mul_arm_func.h b/src/operators/kernel/central-arm-func/mul_arm_func.h
index 60f6bca611a2c053b4e491964cd473de70031a92..8b9dad90a0b02ebf761bcd44fabc18905b056e6e 100644
--- a/src/operators/kernel/central-arm-func/mul_arm_func.h
+++ b/src/operators/kernel/central-arm-func/mul_arm_func.h
@@ -73,13 +73,14 @@ void MulCompute(const MulParam<CPU> &param) {
   }
   if (param.InputX()->type() == typeid(int8_t)) {
     out->mutable_data<int32_t>();
-    math::matmul(x_matrix, false, y_matrix, false, static_cast<float>(1), out,
-                 static_cast<float>(0), false, static_cast<int32_t *>(nullptr));
+    math::matmul<float, int32_t>(x_matrix, false, y_matrix, false,
+                                 static_cast<float>(1), out,
+                                 static_cast<float>(0));
 
   } else {
     out->mutable_data<float>();
-    math::matmul(x_matrix, false, y_matrix, false, static_cast<float>(1), out,
-                 static_cast<float>(0), false, static_cast<float *>(nullptr));
+    math::matmul<float>(x_matrix, false, y_matrix, false, static_cast<float>(1),
+                        out, static_cast<float>(0));
   }
   if (out_dim.size() != 2) {
     out->Resize(out_dim);
diff --git a/test/common/test_gemm_perf.cpp b/test/common/test_gemm_perf.cpp
index c8081e2d47061efdfea8db6c0393dc236d33b3e7..f25a290aef6e228aff0a84d2640486235e0116bf 100644
--- a/test/common/test_gemm_perf.cpp
+++ b/test/common/test_gemm_perf.cpp
@@ -85,16 +85,16 @@ int main() {
   // int8_t without bias
   // warm-up 10 times
   for (int j = 0; j < 10; ++j) {
-    paddle_mobile::operators::math::matmul(
+    paddle_mobile::operators::math::matmul<float, int32_t>(
         aa_int8, false, bb_int8, false, static_cast<float>(1), &cc_int32,
-        static_cast<float>(0), false, static_cast<int32_t*>(nullptr));
+        static_cast<float>(0));
   }
 
   auto time3 = time();
   for (int j = 0; j < 10; ++j) {
-    paddle_mobile::operators::math::matmul(
+    paddle_mobile::operators::math::matmul<float, int32_t>(
         aa_int8, false, bb_int8, false, static_cast<float>(1), &cc_int32,
-        static_cast<float>(0), false, static_cast<int32_t*>(nullptr));
+        static_cast<float>(0));
   }
   auto time4 = time();
   std::cout << "int8_t gemm  cost :" << time_diff(time3, time4) / 10 << "ms\n";
diff --git a/test/operators/test_fusion_conv_add_relu_int8_op.cpp b/test/operators/test_fusion_conv_add_relu_int8_op.cpp
index 8d7067898e98982a506c7ed707800d07aa57884e..add38b34f19061ba56e6ff1f58827671af52c5ee 100644
--- a/test/operators/test_fusion_conv_add_relu_int8_op.cpp
+++ b/test/operators/test_fusion_conv_add_relu_int8_op.cpp
@@ -14,8 +14,8 @@ limitations under the License. */
 
 #ifdef FUSION_CONVADDRELU_INT8_OP
 
-#include <limits>
 #include <iostream>
+#include <limits>
 #include "../test_helper.h"
 #include "../test_include.h"
 #include "operators/fusion_conv_add_relu_int8_op.h"