diff --git a/mace/kernels/benchmark/addn_benchmark.cc b/mace/kernels/benchmark/addn_benchmark.cc
index e735965a665b43324885c95fb41bda3f9af78d50..d717c44e076cf5daf9d2d06aec787400a4242c97 100644
--- a/mace/kernels/benchmark/addn_benchmark.cc
+++ b/mace/kernels/benchmark/addn_benchmark.cc
@@ -32,10 +32,14 @@ static void AddNBenchmark(int iters, int n, int type) {
 
   if (type == DeviceType::CPU) {
     AddNFunctor<DeviceType::CPU, float> addn_functor;
-    addn_functor(inputs, &output[0], n);
+    while (--iters) {
+      addn_functor(inputs, &output[0], n);
+    }
   } else if (type == DeviceType::NEON) {
     AddNFunctor<DeviceType::NEON, float> neon_addn_functor;
-    neon_addn_functor(inputs, &output[0], n);
+    while (--iters) {
+      neon_addn_functor(inputs, &output[0], n);
+    }
   }
 }
 
diff --git a/mace/kernels/benchmark/relu_benchmark.cc b/mace/kernels/benchmark/relu_benchmark.cc
index c384e882e17454bcb91890a28d762bb74f4f6e2d..2d5067b9ae1a5f54bb8d739bb6d9dce07c52012c 100644
--- a/mace/kernels/benchmark/relu_benchmark.cc
+++ b/mace/kernels/benchmark/relu_benchmark.cc
@@ -27,10 +27,14 @@ static void ReluBenchmark(int iters, int n, int type) {
 
   if (type == DeviceType::CPU) {
     ReluFunctor<DeviceType::CPU, float> relu_functor;
-    relu_functor(&input[0], &output[0], n);
+    while (--iters) {
+      relu_functor(&input[0], &output[0], n);
+    }
   } else if (type == DeviceType::NEON) {
     ReluFunctor<DeviceType::NEON, float> neon_relu_functor;
-    neon_relu_functor(&input[0], &output[0], n);
+    while (--iters) {
+      neon_relu_functor(&input[0], &output[0], n);
+    }
   }
 }