Use iters at relu and addn benchmark.

mace/kernels/benchmark/addn_benchmark.cc

@@ -32,11 +32,15 @@ static void AddNBenchmark(int iters, int n, int type) {
 
   if (type == DeviceType::CPU) {
     AddNFunctor<DeviceType::CPU, float> addn_functor;
+    while (--iters) {
       addn_functor(inputs, &output[0], n);
+    }
   } else if (type == DeviceType::NEON) {
     AddNFunctor<DeviceType::NEON, float> neon_addn_functor;
+    while (--iters) {
       neon_addn_functor(inputs, &output[0], n);
     }
+  }
 }
 
 static const int kBenchmarkSize = 10000000;

mace/kernels/benchmark/relu_benchmark.cc

@@ -27,11 +27,15 @@ static void ReluBenchmark(int iters, int n, int type) {
 
   if (type == DeviceType::CPU) {
     ReluFunctor<DeviceType::CPU, float> relu_functor;
+    while (--iters) {
       relu_functor(&input[0], &output[0], n);
+    }
   } else if (type == DeviceType::NEON) {
     ReluFunctor<DeviceType::NEON, float> neon_relu_functor;
+    while (--iters) {
       neon_relu_functor(&input[0], &output[0], n);
     }
+  }
 }
 
 static const int kBenchmarkSize = 10000000;