diff --git a/mace/benchmark/benchmark_model.cc b/mace/benchmark/benchmark_model.cc
index 95b093c1b2ec087085b30fcd1d33633bba2b8abb..38db846b2a5c651e4dead8357ce1753649f6bd65 100644
--- a/mace/benchmark/benchmark_model.cc
+++ b/mace/benchmark/benchmark_model.cc
@@ -2,16 +2,18 @@
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
 
-#include "gflags/gflags.h"
-#include "mace/public/mace.h"
-#include "mace/utils/logging.h"
-#include "mace/benchmark/stat_summarizer.h"
+
+#include <sys/time.h>
 
 #include <cstdlib>
 #include <fstream>
-#include <thread>
 #include <numeric>
-#include <sys/time.h>
+#include <thread>  // NOLINT(build/c++11)
+
+#include "gflags/gflags.h"
+#include "mace/public/mace.h"
+#include "mace/utils/logging.h"
+#include "mace/benchmark/stat_summarizer.h"
 
 namespace mace {
 namespace MACE_MODEL_TAG {
@@ -24,10 +26,11 @@ extern NetDef CreateNet(const unsigned char *model_data);
 
 extern const std::string ModelChecksum();
 
-}
-}
+}  // namespace MACE_MODEL_TAG
+}  // namespace mace
 
 namespace mace {
+namespace benchmark {
 namespace str_util {
 
 std::vector<std::string> Split(const std::string &str, char delims) {
@@ -64,8 +67,6 @@ bool SplitAndParseToInts(const std::string &str,
 
 }  //  namespace str_util
 
-namespace benchmark {
-
 void ParseShape(const std::string &str, std::vector<int64_t> *shape) {
   std::string tmp = str;
   while (!tmp.empty()) {
@@ -96,18 +97,19 @@ inline int64_t NowMicros() {
 
 bool RunInference(MaceEngine *engine,
                   const std::vector<mace::MaceInputInfo> &input_infos,
-                  std::map<std::string, float*> &output_infos,
+                  std::map<std::string, float*> *output_infos,
                   StatSummarizer *summarizer,
                   int64_t *inference_time_us) {
+  MACE_CHECK_NOTNULL(output_infos);
   RunMetadata run_metadata;
   RunMetadata *run_metadata_ptr = nullptr;
   if (summarizer) {
     run_metadata_ptr = &run_metadata;
   }
-  if (input_infos.size() == 1 && output_infos.size() == 1) {
+  if (input_infos.size() == 1 && output_infos->size() == 1) {
     const int64_t start_time = NowMicros();
     bool s = engine->Run(input_infos[0].data, input_infos[0].shape,
-                         output_infos.begin()->second, run_metadata_ptr);
+                         output_infos->begin()->second, run_metadata_ptr);
     const int64_t end_time = NowMicros();
 
     if (!s) {
@@ -117,7 +119,7 @@ bool RunInference(MaceEngine *engine,
     *inference_time_us = end_time - start_time;
   } else {
     const int64_t start_time = NowMicros();
-    bool s = engine->Run(input_infos, output_infos, run_metadata_ptr);
+    bool s = engine->Run(input_infos, *output_infos, run_metadata_ptr);
     const int64_t end_time = NowMicros();
 
     if (!s) {
@@ -136,13 +138,14 @@ bool RunInference(MaceEngine *engine,
 
 bool Run(MaceEngine *engine,
          const std::vector<mace::MaceInputInfo> &input_infos,
-         std::map<std::string, float*> &output_infos,
+         std::map<std::string, float*> *output_infos,
          StatSummarizer *summarizer,
          int num_runs,
          double max_time_sec,
          int64_t sleep_sec,
          int64_t *total_time_us,
          int64_t *actual_num_runs) {
+  MACE_CHECK_NOTNULL(output_infos);
   *total_time_us = 0;
 
   LOG(INFO) << "Running benchmark for max " << num_runs << " iterators, max "
@@ -156,7 +159,8 @@ bool Run(MaceEngine *engine,
   bool util_max_time = (num_runs <= 0);
   for (int i = 0; util_max_time || i < num_runs; ++i) {
     int64_t inference_time_us = 0;
-    bool s = RunInference(engine, input_infos, output_infos, summarizer, &inference_time_us);
+    bool s = RunInference(engine, input_infos, output_infos,
+                          summarizer, &inference_time_us);
     stat.UpdateStat(inference_time_us);
     (*total_time_us) += inference_time_us;
     ++(*actual_num_runs);
@@ -183,15 +187,18 @@ bool Run(MaceEngine *engine,
 }
 
 DEFINE_string(device, "CPU", "Device [CPU|OPENCL]");
-DEFINE_string(input_node, "input_node0,input_node1", "input nodes, separated by comma");
-DEFINE_string(output_node, "output_node0,output_node1", "output nodes, separated by comma");
+DEFINE_string(input_node, "input_node0,input_node1",
+              "input nodes, separated by comma");
+DEFINE_string(output_node, "output_node0,output_node1",
+              "output nodes, separated by comma");
 DEFINE_string(input_shape, "", "input shape, separated by colon and comma");
 DEFINE_string(output_shape, "", "output shape, separated by colon and comma");
 DEFINE_string(input_file, "", "input file name");
 DEFINE_int32(max_num_runs, 100, "number of runs max");
 DEFINE_string(max_time, "10.0", "length to run max");
 DEFINE_string(inference_delay, "-1", "delay between runs in seconds");
-DEFINE_string(inter_benchmark_delay, "-1", "delay between benchmarks in seconds");
+DEFINE_string(inter_benchmark_delay, "-1",
+              "delay between benchmarks in seconds");
 DEFINE_string(benchmark_name, "", "benchmark name");
 DEFINE_bool(show_run_order, true, "whether to list stats by run order");
 DEFINE_int32(run_order_limit, 0, "how many items to show by run order");
@@ -203,15 +210,18 @@ DEFINE_bool(show_type, true, "whether to list stats by op type");
 DEFINE_bool(show_summary, true, "whether to show a summary of the stats");
 DEFINE_bool(show_flops, true, "whether to estimate the model's FLOPs");
 DEFINE_int32(warmup_runs, 1, "how many runs to initialize model");
-DEFINE_string(model_data_file, "", "model data file name, used when EMBED_MODEL_DATA set to 0");
+DEFINE_string(model_data_file, "",
+              "model data file name, used when EMBED_MODEL_DATA set to 0");
 DEFINE_string(gpu_type, "ADRENO", "ADRENO/MALI");
 DEFINE_int32(gpu_perf_hint, 2, "0:DEFAULT/1:LOW/2:NORMAL/3:HIGH");
 DEFINE_int32(gpu_priority_hint, 1, "0:DEFAULT/1:LOW/2:NORMAL/3:HIGH");
 DEFINE_int32(omp_num_threads, 8, "num of openmp threads");
-DEFINE_int32(cpu_power_option, 0, "0:DEFAULT/1:HIGH_PERFORMANCE/2:BATTERY_SAVE");
+DEFINE_int32(cpu_power_option, 0,
+             "0:DEFAULT/1:HIGH_PERFORMANCE/2:BATTERY_SAVE");
 
 int Main(int argc, char **argv) {
-  MACE_CHECK(FLAGS_device != "HEXAGON", "Model benchmark tool do not support DSP.");
+  MACE_CHECK(FLAGS_device != "HEXAGON",
+             "Model benchmark tool do not support DSP.");
   gflags::SetUsageMessage("some usage message");
   gflags::ParseCommandLineFlags(&argc, &argv, true);
 
@@ -228,12 +238,14 @@ int Main(int argc, char **argv) {
   LOG(INFO) << "output shapes: [" << FLAGS_output_shape << "]";
   LOG(INFO) << "Warmup runs: [" << FLAGS_warmup_runs << "]";
   LOG(INFO) << "Num runs: [" << FLAGS_max_num_runs << "]";
-  LOG(INFO) << "Inter-inference delay (seconds): [" << FLAGS_inference_delay << "]";
-  LOG(INFO) << "Inter-benchmark delay (seconds): [" << FLAGS_inter_benchmark_delay << "]";
+  LOG(INFO) << "Inter-inference delay (seconds): ["
+            << FLAGS_inference_delay << "]";
+  LOG(INFO) << "Inter-benchmark delay (seconds): ["
+            << FLAGS_inter_benchmark_delay << "]";
 
-  const long int inter_inference_sleep_seconds =
+  const int64_t inter_inference_sleep_seconds =
       std::strtol(FLAGS_inference_delay.c_str(), nullptr, 10);
-  const long int inter_benchmark_sleep_seconds =
+  const int64_t inter_benchmark_sleep_seconds =
       std::strtol(FLAGS_inter_benchmark_delay.c_str(), nullptr, 10);
   const double max_benchmark_time_seconds =
       std::strtod(FLAGS_max_time.c_str(), nullptr);
@@ -252,7 +264,7 @@ int Main(int argc, char **argv) {
   stats.reset(new StatSummarizer(stats_options));
 
   DeviceType device_type = CPU;
-  if(FLAGS_device == "OPENCL") {
+  if (FLAGS_device == "OPENCL") {
     device_type = OPENCL;
   }
 
@@ -264,17 +276,20 @@ int Main(int argc, char **argv) {
         gpu_type,
         static_cast<GPUPerfHint>(FLAGS_gpu_perf_hint),
         static_cast<GPUPriorityHint>(FLAGS_gpu_priority_hint));
-  }
-  else if (device_type == CPU) {
+  } else if (device_type == CPU) {
     mace::ConfigOmpThreadsAndAffinity(
         FLAGS_omp_num_threads,
         static_cast<CPUPowerOption>(FLAGS_cpu_power_option));
   }
 
-  std::vector<std::string> input_names = str_util::Split(FLAGS_input_node, ',');
-  std::vector<std::string> output_names = str_util::Split(FLAGS_output_node, ',');
-  std::vector<std::string> input_shapes = str_util::Split(FLAGS_input_shape, ':');
-  std::vector<std::string> output_shapes = str_util::Split(FLAGS_output_shape, ':');
+  std::vector<std::string> input_names =
+      str_util::Split(FLAGS_input_node, ',');
+  std::vector<std::string> output_names =
+      str_util::Split(FLAGS_output_node, ',');
+  std::vector<std::string> input_shapes =
+      str_util::Split(FLAGS_input_shape, ':');
+  std::vector<std::string> output_shapes =
+      str_util::Split(FLAGS_output_shape, ':');
 
   const size_t input_count = input_shapes.size();
   const size_t output_count = output_shapes.size();
@@ -298,10 +313,12 @@ int Main(int argc, char **argv) {
 
   for (size_t i = 0; i < input_count; ++i) {
     int64_t input_size = std::accumulate(input_shape_vec[i].begin(),
-                                         input_shape_vec[i].end(), 1, std::multiplies<int64_t>());
+                                         input_shape_vec[i].end(), 1,
+                                         std::multiplies<int64_t>());
     input_datas[i].reset(new float[input_size]);
     // load input
-    std::ifstream in_file(FLAGS_input_file + "_" + FormatName(input_names[i]), std::ios::in | std::ios::binary);
+    std::ifstream in_file(FLAGS_input_file + "_" + FormatName(input_names[i]),
+                          std::ios::in | std::ios::binary);
     if (in_file.is_open()) {
       in_file.read(reinterpret_cast<char *>(input_datas[i].get()),
                    input_size * sizeof(float));
@@ -317,7 +334,8 @@ int Main(int argc, char **argv) {
   }
   for (size_t i = 0; i < output_count; ++i) {
     int64_t output_size = std::accumulate(output_shape_vec[i].begin(),
-                                          output_shape_vec[i].end(), 1, std::multiplies<int64_t>());
+                                          output_shape_vec[i].end(), 1,
+                                          std::multiplies<int64_t>());
     output_datas[i].reset(new float[output_size]);
     output_infos[output_names[i]] = output_datas[i].get();
   }
@@ -328,7 +346,8 @@ int Main(int argc, char **argv) {
   if (input_count == 1 && output_count == 1) {
     engine_ptr.reset(new mace::MaceEngine(&net_def, device_type));
   } else {
-    engine_ptr.reset(new mace::MaceEngine(&net_def, device_type, input_names, output_names));
+    engine_ptr.reset(new mace::MaceEngine(&net_def, device_type,
+                                          input_names, output_names));
   }
   if (device_type == DeviceType::OPENCL) {
     mace::MACE_MODEL_TAG::UnloadModelData(model_data);
@@ -340,7 +359,8 @@ int Main(int argc, char **argv) {
   int64_t num_warmup_runs = 0;
   if (FLAGS_warmup_runs > 0) {
     bool status =
-        Run(engine_ptr.get(), input_infos, output_infos, nullptr, FLAGS_warmup_runs, -1.0,
+        Run(engine_ptr.get(), input_infos, &output_infos, nullptr,
+            FLAGS_warmup_runs, -1.0,
             inter_inference_sleep_seconds, &warmup_time_us, &num_warmup_runs);
     if (!status) {
       LOG(ERROR) << "Failed at warm up run";
@@ -354,7 +374,7 @@ int Main(int argc, char **argv) {
   int64_t no_stat_time_us = 0;
   int64_t no_stat_runs = 0;
   bool status =
-      Run(engine_ptr.get(), input_infos, output_infos,
+      Run(engine_ptr.get(), input_infos, &output_infos,
           nullptr, FLAGS_max_num_runs, max_benchmark_time_seconds,
           inter_inference_sleep_seconds, &no_stat_time_us, &no_stat_runs);
   if (!status) {
@@ -363,7 +383,7 @@ int Main(int argc, char **argv) {
 
   int64_t stat_time_us = 0;
   int64_t stat_runs = 0;
-  status = Run(engine_ptr.get(), input_infos, output_infos,
+  status = Run(engine_ptr.get(), input_infos, &output_infos,
                stats.get(), FLAGS_max_num_runs, max_benchmark_time_seconds,
                inter_inference_sleep_seconds, &stat_time_us, &stat_runs);
   if (!status) {
@@ -372,8 +392,8 @@ int Main(int argc, char **argv) {
 
   LOG(INFO) << "Average inference timings in us: "
             << "Warmup: "
-            << (FLAGS_warmup_runs > 0 ? warmup_time_us / FLAGS_warmup_runs : 0) << ", "
-            << "no stats: " << no_stat_time_us / no_stat_runs << ", "
+            << (FLAGS_warmup_runs > 0 ? warmup_time_us / FLAGS_warmup_runs : 0)
+            << ", " << "no stats: " << no_stat_time_us / no_stat_runs << ", "
             << "with stats: " << stat_time_us / stat_runs;
 
   stats->PrintOperatorStats();
@@ -381,7 +401,7 @@ int Main(int argc, char **argv) {
   return 0;
 }
 
-}  //  namespace benchmark
-}  //  namespace mace
+}  // namespace benchmark
+}  // namespace mace
 
 int main(int argc, char **argv) { mace::benchmark::Main(argc, argv); }
diff --git a/mace/benchmark/model_throughput_test.cc b/mace/benchmark/model_throughput_test.cc
index 48e622cb7b79b026b89ad6eb1fb4349056b2d9e8..bc2be7a6c61f4f7f796884cebed17d340dd2ebe1 100644
--- a/mace/benchmark/model_throughput_test.cc
+++ b/mace/benchmark/model_throughput_test.cc
@@ -19,16 +19,13 @@
 #include <fstream>
 #include <iostream>
 #include <numeric>
-#include <thread>
+#include <thread>  // NOLINT(build/c++11)
 
 #include "gflags/gflags.h"
 #include "mace/public/mace.h"
 #include "mace/utils/env_time.h"
 #include "mace/utils/logging.h"
 
-using namespace std;
-using namespace mace;
-
 namespace mace {
 
 #ifdef MACE_CPU_MODEL_TAG
@@ -73,15 +70,15 @@ extern const std::string ModelChecksum();
 }  // namespace MACE_DSP_MODEL_TAG
 #endif
 
-}  // namespace mace
+namespace benchmark {
 
-void ParseShape(const string &str, vector<int64_t> *shape) {
-  string tmp = str;
+void ParseShape(const std::string &str, std::vector<int64_t> *shape) {
+  std::string tmp = str;
   while (!tmp.empty()) {
     int dim = atoi(tmp.data());
     shape->push_back(dim);
     size_t next_offset = tmp.find(",");
-    if (next_offset == string::npos) {
+    if (next_offset == std::string::npos) {
       break;
     } else {
       tmp = tmp.substr(next_offset + 1);
@@ -89,7 +86,7 @@ void ParseShape(const string &str, vector<int64_t> *shape) {
   }
 }
 
-DeviceType ParseDeviceType(const string &device_str) {
+DeviceType ParseDeviceType(const std::string &device_str) {
   if (device_str.compare("CPU") == 0) {
     return DeviceType::CPU;
   } else if (device_str.compare("NEON") == 0) {
@@ -111,20 +108,23 @@ DEFINE_string(gpu_model_data_file, "", "gpu model data file name");
 DEFINE_string(dsp_model_data_file, "", "dsp model data file name");
 DEFINE_int32(run_seconds, 10, "run seconds");
 
-int main(int argc, char **argv) {
+int Main(int argc, char **argv) {
   gflags::SetUsageMessage("some usage message");
   gflags::ParseCommandLineFlags(&argc, &argv, true);
 
   LOG(INFO) << "mace version: " << MaceVersion();
   LOG(INFO) << "mace git version: " << MaceGitVersion();
 #ifdef MACE_CPU_MODEL_TAG
-  LOG(INFO) << "cpu model checksum: " << mace::MACE_CPU_MODEL_TAG::ModelChecksum();
+  LOG(INFO) << "cpu model checksum: "
+            << mace::MACE_CPU_MODEL_TAG::ModelChecksum();
 #endif
 #ifdef MACE_GPU_MODEL_TAG
-  LOG(INFO) << "gpu model checksum: " << mace::MACE_GPU_MODEL_TAG::ModelChecksum();
+  LOG(INFO) << "gpu model checksum: "
+            << mace::MACE_GPU_MODEL_TAG::ModelChecksum();
 #endif
 #ifdef MACE_DSP_MODEL_TAG
-  LOG(INFO) << "dsp model checksum: " << mace::MACE_DSP_MODEL_TAG::ModelChecksum();
+  LOG(INFO) << "dsp model checksum: "
+            << mace::MACE_DSP_MODEL_TAG::ModelChecksum();
 #endif
   LOG(INFO) << "input_shape: " << FLAGS_input_shape;
   LOG(INFO) << "output_shape: " << FLAGS_output_shape;
@@ -134,8 +134,8 @@ int main(int argc, char **argv) {
   LOG(INFO) << "dsp_model_data_file: " << FLAGS_dsp_model_data_file;
   LOG(INFO) << "run_seconds: " << FLAGS_run_seconds;
 
-  vector<int64_t> input_shape_vec;
-  vector<int64_t> output_shape_vec;
+  std::vector<int64_t> input_shape_vec;
+  std::vector<int64_t> output_shape_vec;
   ParseShape(FLAGS_input_shape, &input_shape_vec);
   ParseShape(FLAGS_output_shape, &output_shape_vec);
 
@@ -151,7 +151,7 @@ int main(int argc, char **argv) {
   std::unique_ptr<float[]> dsp_output_data(new float[output_size]);
 
   // load input
-  ifstream in_file(FLAGS_input_file, ios::in | ios::binary);
+  std::ifstream in_file(FLAGS_input_file, std::ios::in | std::ios::binary);
   if (in_file.is_open()) {
     in_file.read(reinterpret_cast<char *>(input_data.get()),
                  input_size * sizeof(float));
@@ -166,7 +166,8 @@ int main(int argc, char **argv) {
   /* --------------------- CPU init ----------------------- */
   LOG(INFO) << "Load & init cpu model and warm up";
   const unsigned char *cpu_model_data =
-      mace::MACE_CPU_MODEL_TAG::LoadModelData(FLAGS_cpu_model_data_file.c_str());
+      mace::MACE_CPU_MODEL_TAG::LoadModelData(
+      FLAGS_cpu_model_data_file.c_str());
   NetDef cpu_net_def = mace::MACE_CPU_MODEL_TAG::CreateNet(cpu_model_data);
 
   mace::MaceEngine cpu_engine(&cpu_net_def, DeviceType::CPU);
@@ -182,7 +183,8 @@ int main(int argc, char **argv) {
   /* --------------------- GPU init ----------------------- */
   LOG(INFO) << "Load & init gpu model and warm up";
   const unsigned char *gpu_model_data =
-      mace::MACE_GPU_MODEL_TAG::LoadModelData(FLAGS_gpu_model_data_file.c_str());
+      mace::MACE_GPU_MODEL_TAG::LoadModelData(
+      FLAGS_gpu_model_data_file.c_str());
   NetDef gpu_net_def = mace::MACE_GPU_MODEL_TAG::CreateNet(gpu_model_data);
 
   mace::MaceEngine gpu_engine(&gpu_net_def, DeviceType::OPENCL);
@@ -199,7 +201,8 @@ int main(int argc, char **argv) {
   /* --------------------- DSP init ----------------------- */
   LOG(INFO) << "Load & init dsp model and warm up";
   const unsigned char *dsp_model_data =
-      mace::MACE_DSP_MODEL_TAG::LoadModelData(FLAGS_gpu_model_data_file.c_str());
+      mace::MACE_DSP_MODEL_TAG::LoadModelData(
+      FLAGS_gpu_model_data_file.c_str());
   NetDef dsp_net_def = mace::MACE_DSP_MODEL_TAG::CreateNet(dsp_model_data);
 
   mace::MaceEngine dsp_engine(&dsp_net_def, DeviceType::HEXAGON);
@@ -278,4 +281,11 @@ int main(int argc, char **argv) {
 #endif
 
   LOG(INFO) << "Total throughput: " << total_throughput << " f/s";
+
+  return 0;
 }
+
+}  // namespace benchmark
+}  // namespace mace
+
+int main(int argc, char **argv) { mace::benchmark::Main(argc, argv); }
diff --git a/mace/benchmark/stat_summarizer.cc b/mace/benchmark/stat_summarizer.cc
index ca8c5ec3db112ff4f6b3ea571b7b3a01c327bb2e..10a89d807d356cc5714e16520f0a3b3d200d9273 100644
--- a/mace/benchmark/stat_summarizer.cc
+++ b/mace/benchmark/stat_summarizer.cc
@@ -3,14 +3,18 @@
 //
 
 #include "mace/benchmark/stat_summarizer.h"
-#include "mace/public/mace.h"
-#include "mace/utils/logging.h"
 
 #include <iomanip>
-#include <queue>
 #include <iostream>
+#include <queue>
+#include <utility>
+
+#include "mace/public/mace.h"
+#include "mace/utils/logging.h"
+
 
 namespace mace {
+namespace benchmark {
 
 StatSummarizer::StatSummarizer(const StatSummarizerOptions &options)
     : options_(options) {}
@@ -90,7 +94,7 @@ std::string StatSummarizer::HeaderString(const std::string &title) const {
   stream << "============================== " << title
          << " ==============================" << std::endl;
 
-  InitField(stream, 14) << "[node type]";
+  InitField(stream, 24) << "[node type]";
   InitField(stream, 9) << "[start]";
   InitField(stream, 9) << "[first]";
   InitField(stream, 9) << "[avg ms]";
@@ -114,7 +118,7 @@ std::string StatSummarizer::ColumnString(const StatSummarizer::Detail &detail,
   const int64_t times_called = detail.times_called / num_runs();
 
   std::stringstream stream;
-  InitField(stream, 14) << detail.type;
+  InitField(stream, 24) << detail.type;
   InitField(stream, 9) << start_ms;
   InitField(stream, 9) << first_time_ms;
   InitField(stream, 9) << avg_time_ms;
@@ -224,7 +228,7 @@ std::string StatSummarizer::GetStatsByNodeType() const {
                     std::pair<std::string, int64_t>(node_type.first, mem_used));
   }
 
-  InitField(stream, 14) << "[Node type]";
+  InitField(stream, 24) << "[Node type]";
   InitField(stream, 9) << "[count]";
   InitField(stream, 10) << "[avg ms]";
   InitField(stream, 11) << "[avg %]";
@@ -248,7 +252,7 @@ std::string StatSummarizer::GetStatsByNodeType() const {
         ((entry.first / static_cast<float>(accumulated_us)) * 100.0f);
     cdf += percentage;
 
-    InitField(stream, 14) << node_type;
+    InitField(stream, 24) << node_type;
     InitField(stream, 9) << node_type_map_count[node_type];
     InitField(stream, 10) << time_per_run_ms;
     InitField(stream, 10) << percentage << "%";
@@ -317,4 +321,5 @@ void StatSummarizer::PrintOperatorStats() const {
   }
 }
 
+}  // namespace benchmark
 }  // namespace mace
diff --git a/mace/benchmark/stat_summarizer.h b/mace/benchmark/stat_summarizer.h
index 368b1e30299b77467b847396eeb20df586895618..8d2b99ad0d944fd8ff49d65db47bcc0fda494d87 100644
--- a/mace/benchmark/stat_summarizer.h
+++ b/mace/benchmark/stat_summarizer.h
@@ -2,10 +2,11 @@
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
 
-#ifndef MACE_TOOLS_BENCHMARK_STAT_SUMMARIZER_H_
-#define MACE_TOOLS_BENCHMARK_STAT_SUMMARIZER_H_
+#ifndef MACE_BENCHMARK_STAT_SUMMARIZER_H_
+#define MACE_BENCHMARK_STAT_SUMMARIZER_H_
 
 #include <stdlib.h>
+#include <algorithm>
 #include <cmath>
 #include <limits>
 #include <map>
@@ -17,6 +18,8 @@ namespace mace {
 
 class RunMetadata;
 
+namespace benchmark {
+
 template <typename ValueType, typename HighPrecisionValueType = double>
 class Stat {
  public:
@@ -194,6 +197,7 @@ class StatSummarizer {
   StatSummarizerOptions options_;
 };
 
-}  //  namespace mace
+}  // namespace benchmark
+}  // namespace mace
 
-#endif  // MACE_TOOLS_BENCHMARK_STAT_SUMMARIZER_H_
+#endif  // MACE_BENCHMARK_STAT_SUMMARIZER_H_
diff --git a/mace/codegen/opencl/opencl_compiled_program.cc b/mace/codegen/opencl/opencl_compiled_program.cc
index 35c5d6417e8e11205799f20c1d9d533eb13271a7..62fbad33022b92d4d72a7f407cca18c5e4073fe3 100644
--- a/mace/codegen/opencl/opencl_compiled_program.cc
+++ b/mace/codegen/opencl/opencl_compiled_program.cc
@@ -13,4 +13,4 @@ namespace mace {
 extern const std::map<std::string, std::vector<unsigned char>>
     kCompiledProgramMap = {};
 
-}  // namespace
+}  // namespace mace
diff --git a/mace/codegen/tuning/tuning_params.cc b/mace/codegen/tuning/tuning_params.cc
index 12b5199bc2dcff7dab6bbd3fb68e0c9c7addd0f6..8a8b7132a26a150d14ff9504b13aa668fc109285 100644
--- a/mace/codegen/tuning/tuning_params.cc
+++ b/mace/codegen/tuning/tuning_params.cc
@@ -13,4 +13,4 @@ namespace mace {
 extern const std::map<std::string, std::vector<unsigned int>>
     kTuningParamsData = {};
 
-}  // namespace
+}  // namespace mace
diff --git a/mace/examples/mace_run.cc b/mace/examples/mace_run.cc
index 68e986fb29470b61ebc9009fed7589c360a26f31..373b2455a3cba8e70911c652d04d94fd270834fb 100644
--- a/mace/examples/mace_run.cc
+++ b/mace/examples/mace_run.cc
@@ -16,8 +16,6 @@
  */
 #include <malloc.h>
 #include <stdint.h>
-#include <sys/time.h>
-#include <time.h>
 #include <cstdlib>
 #include <fstream>
 #include <iostream>
@@ -28,9 +26,6 @@
 #include "mace/utils/env_time.h"
 #include "mace/utils/logging.h"
 
-using namespace std;
-using namespace mace;
-
 namespace mace {
 namespace MACE_MODEL_TAG {
 
@@ -45,6 +40,8 @@ extern const std::string ModelChecksum();
 }  // namespace MACE_MODEL_TAG
 }  // namespace mace
 
+namespace mace {
+namespace examples {
 
 namespace str_util {
 
@@ -65,13 +62,13 @@ std::vector<std::string> Split(const std::string &str, char delims) {
 
 }  // namespace str_util
 
-void ParseShape(const string &str, vector<int64_t> *shape) {
-  string tmp = str;
+void ParseShape(const std::string &str, std::vector<int64_t> *shape) {
+  std::string tmp = str;
   while (!tmp.empty()) {
     int dim = atoi(tmp.data());
     shape->push_back(dim);
     size_t next_offset = tmp.find(",");
-    if (next_offset == string::npos) {
+    if (next_offset == std::string::npos) {
       break;
     } else {
       tmp = tmp.substr(next_offset + 1);
@@ -87,7 +84,7 @@ std::string FormatName(const std::string input) {
   return res;
 }
 
-DeviceType ParseDeviceType(const string &device_str) {
+DeviceType ParseDeviceType(const std::string &device_str) {
   if (device_str.compare("CPU") == 0) {
     return DeviceType::CPU;
   } else if (device_str.compare("NEON") == 0) {
@@ -101,7 +98,7 @@ DeviceType ParseDeviceType(const string &device_str) {
   }
 }
 
-GPUType ParseGPUType(const string &gpu_type_str) {
+GPUType ParseGPUType(const std::string &gpu_type_str) {
   if (gpu_type_str.compare("ADRENO") == 0) {
     return GPUType::ADRENO;
   } else if (gpu_type_str.compare("MALI") == 0) {
@@ -158,12 +155,18 @@ struct mallinfo LogMallinfoChange(struct mallinfo prev) {
   return curr;
 }
 
-DEFINE_string(input_node, "input_node0,input_node1", "input nodes, separated by comma");
-DEFINE_string(input_shape, "1,224,224,3:1,1,1,10", "input shapes, separated by colon and comma");
-DEFINE_string(output_node, "output_node0,output_node1", "output nodes, separated by comma");
-DEFINE_string(output_shape, "1,224,224,2:1,1,1,10", "output shapes, separated by colon and comma");
-DEFINE_string(input_file, "", "input file name | input file prefix for multiple inputs.");
-DEFINE_string(output_file, "", "output file name | output file prefix for multiple outputs");
+DEFINE_string(input_node, "input_node0,input_node1",
+              "input nodes, separated by comma");
+DEFINE_string(input_shape, "1,224,224,3:1,1,1,10",
+              "input shapes, separated by colon and comma");
+DEFINE_string(output_node, "output_node0,output_node1",
+              "output nodes, separated by comma");
+DEFINE_string(output_shape, "1,224,224,2:1,1,1,10",
+              "output shapes, separated by colon and comma");
+DEFINE_string(input_file, "",
+              "input file name | input file prefix for multiple inputs.");
+DEFINE_string(output_file, "",
+              "output file name | output file prefix for multiple outputs");
 DEFINE_string(model_data_file, "",
               "model data file name, used when EMBED_MODEL_DATA set to 0");
 DEFINE_string(device, "OPENCL", "CPU/NEON/OPENCL/HEXAGON");
@@ -174,7 +177,8 @@ DEFINE_string(gpu_type, "ADRENO", "ADRENO/MALI");
 DEFINE_int32(gpu_perf_hint, 2, "0:DEFAULT/1:LOW/2:NORMAL/3:HIGH");
 DEFINE_int32(gpu_priority_hint, 1, "0:DEFAULT/1:LOW/2:NORMAL/3:HIGH");
 DEFINE_int32(omp_num_threads, 8, "num of openmp threads");
-DEFINE_int32(cpu_power_option, 0, "0:DEFAULT/1:HIGH_PERFORMANCE/2:BATTERY_SAVE");
+DEFINE_int32(cpu_power_option, 0,
+             "0:DEFAULT/1:HIGH_PERFORMANCE/2:BATTERY_SAVE");
 
 bool SingleInputAndOutput(const std::vector<int64_t> &input_shape,
                           const std::vector<int64_t> &output_shape) {
@@ -197,8 +201,7 @@ bool SingleInputAndOutput(const std::vector<int64_t> &input_shape,
         gpu_type,
         static_cast<GPUPerfHint>(FLAGS_gpu_perf_hint),
         static_cast<GPUPriorityHint>(FLAGS_gpu_priority_hint));
-  }
-  else if (device_type == DeviceType::CPU) {
+  } else if (device_type == DeviceType::CPU) {
     mace::ConfigOmpThreadsAndAffinity(
         FLAGS_omp_num_threads,
         static_cast<CPUPowerOption>(FLAGS_cpu_power_option));
@@ -227,7 +230,8 @@ bool SingleInputAndOutput(const std::vector<int64_t> &input_shape,
   std::unique_ptr<float[]> output_data(new float[output_size]);
 
   // load input
-  ifstream in_file(FLAGS_input_file + "_" + FormatName(FLAGS_input_node), ios::in | ios::binary);
+  std::ifstream in_file(FLAGS_input_file + "_" + FormatName(FLAGS_input_node),
+                        std::ios::in | std::ios::binary);
   if (in_file.is_open()) {
     in_file.read(reinterpret_cast<char *>(input_data.get()),
                  input_size * sizeof(float));
@@ -262,7 +266,7 @@ bool SingleInputAndOutput(const std::vector<int64_t> &input_shape,
     if (output_data != nullptr) {
       std::string
         output_name = FLAGS_output_file + "_" + FormatName(FLAGS_output_node);
-      ofstream out_file(output_name, ios::binary);
+      std::ofstream out_file(output_name, std::ios::binary);
       out_file.write((const char *) (output_data.get()),
                      output_size * sizeof(float));
       out_file.flush();
@@ -279,10 +283,11 @@ bool SingleInputAndOutput(const std::vector<int64_t> &input_shape,
   return true;
 }
 
-bool MultipleInputOrOutput(const std::vector<std::string> &input_names,
-                           const std::vector<std::vector<int64_t>> &input_shapes,
-                           const std::vector<std::string> &output_names,
-                           const std::vector<std::vector<int64_t>> &output_shapes) {
+bool MultipleInputOrOutput(
+    const std::vector<std::string> &input_names,
+    const std::vector<std::vector<int64_t>> &input_shapes,
+    const std::vector<std::string> &output_names,
+    const std::vector<std::vector<int64_t>> &output_shapes) {
   // load model
   int64_t t0 = NowMicros();
   const unsigned char *model_data =
@@ -302,8 +307,7 @@ bool MultipleInputOrOutput(const std::vector<std::string> &input_names,
         gpu_type,
         static_cast<GPUPerfHint>(FLAGS_gpu_perf_hint),
         static_cast<GPUPriorityHint>(FLAGS_gpu_priority_hint));
-  }
-  else if (device_type == DeviceType::CPU) {
+  } else if (device_type == DeviceType::CPU) {
     mace::ConfigOmpThreadsAndAffinity(
         FLAGS_omp_num_threads,
         static_cast<CPUPowerOption>(FLAGS_cpu_power_option));
@@ -333,7 +337,8 @@ bool MultipleInputOrOutput(const std::vector<std::string> &input_names,
                         std::multiplies<int64_t>());
     input_datas[i].reset(new float[input_size]);
     // load input
-    ifstream in_file(FLAGS_input_file + "_" + FormatName(input_names[i]), ios::in | ios::binary);
+    std::ifstream in_file(FLAGS_input_file + "_" + FormatName(input_names[i]),
+                          std::ios::in | std::ios::binary);
     if (in_file.is_open()) {
       in_file.read(reinterpret_cast<char *>(input_datas[i].get()),
                    input_size * sizeof(float));
@@ -377,8 +382,9 @@ bool MultipleInputOrOutput(const std::vector<std::string> &input_names,
   }
 
   for (size_t i = 0; i < output_count; ++i) {
-    std::string output_name = FLAGS_output_file + "_" + FormatName(output_names[i]);
-    ofstream out_file(output_name, ios::binary);
+    std::string output_name = FLAGS_output_file + "_"
+        + FormatName(output_names[i]);
+    std::ofstream out_file(output_name, std::ios::binary);
     int64_t output_size =
         std::accumulate(output_shapes[i].begin(), output_shapes[i].end(), 1,
                         std::multiplies<int64_t>());
@@ -395,7 +401,7 @@ bool MultipleInputOrOutput(const std::vector<std::string> &input_names,
   return true;
 }
 
-int main(int argc, char **argv) {
+int Main(int argc, char **argv) {
   gflags::SetUsageMessage("some usage message");
   gflags::ParseCommandLineFlags(&argc, &argv, true);
 
@@ -419,14 +425,17 @@ int main(int argc, char **argv) {
   LOG(INFO) << "cpu_power_option: " << FLAGS_cpu_power_option;
 
   std::vector<std::string> input_names = str_util::Split(FLAGS_input_node, ',');
-  std::vector<std::string> output_names = str_util::Split(FLAGS_output_node, ',');
-  std::vector<std::string> input_shapes = str_util::Split(FLAGS_input_shape, ':');
-  std::vector<std::string> output_shapes = str_util::Split(FLAGS_output_shape, ':');
+  std::vector<std::string> output_names =
+      str_util::Split(FLAGS_output_node, ',');
+  std::vector<std::string> input_shapes =
+      str_util::Split(FLAGS_input_shape, ':');
+  std::vector<std::string> output_shapes =
+      str_util::Split(FLAGS_output_shape, ':');
 
   const size_t input_count = input_shapes.size();
   const size_t output_count = output_shapes.size();
-  std::vector<vector<int64_t>> input_shape_vec(input_count);
-  std::vector<vector<int64_t>> output_shape_vec(output_count);
+  std::vector<std::vector<int64_t>> input_shape_vec(input_count);
+  std::vector<std::vector<int64_t>> output_shape_vec(output_count);
   for (size_t i = 0; i < input_count; ++i) {
     ParseShape(input_shapes[i], &input_shape_vec[i]);
   }
@@ -447,9 +456,14 @@ int main(int argc, char **argv) {
                                   output_shape_vec);
     }
   }
-  if(ret) {
+  if (ret) {
     return 0;
   } else {
     return -1;
   }
 }
+
+}  // namespace examples
+}  // namespace mace
+
+int main(int argc, char **argv) { mace::examples::Main(argc, argv); }
diff --git a/mace/public/mace.h b/mace/public/mace.h
index 3cf5f2973f313022230241aa5edfd11507ddd50f..725d262660f88d2a0ec046dc9d55ada2f836d05a 100644
--- a/mace/public/mace.h
+++ b/mace/public/mace.h
@@ -2,8 +2,8 @@
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
 
-#ifndef MACE_CORE_MACE_H_
-#define MACE_CORE_MACE_H_
+#ifndef MACE_PUBLIC_MACE_H_
+#define MACE_PUBLIC_MACE_H_
 
 #include <cstdint>
 #include <map>
@@ -169,7 +169,7 @@ class NodeInput {
 class OutputShape {
  public:
   OutputShape();
-  OutputShape(const std::vector<int64_t> &dims);
+  OutputShape(const std::vector<int64_t> &dims);  // NOLINT(runtime/explicit)
   void CopyFrom(const OutputShape &from);
 
  public:
@@ -409,9 +409,10 @@ class MaceEngine {
            float *output,
            RunMetadata *run_metadata);
   // Multiple input or output
-  bool Run(const std::vector<MaceInputInfo> &input,
-           std::map<std::string, float *> &output,
-           RunMetadata *run_metadata = nullptr);
+  bool Run(
+      const std::vector<MaceInputInfo> &input,
+      std::map<std::string, float *> &output,  // NOLINT(runtime/references)
+      RunMetadata *run_metadata = nullptr);
   MaceEngine(const MaceEngine &) = delete;
   MaceEngine &operator=(const MaceEngine &) = delete;
 
@@ -425,4 +426,4 @@ class MaceEngine {
 
 }  // namespace mace
 
-#endif  // MACE_CORE_MACE_H_
+#endif  // MACE_PUBLIC_MACE_H_
diff --git a/mace/utils/command_line_flags.cc b/mace/utils/command_line_flags.cc
index 83f8be3cd544e52fa441273bf60d4a0ccd7f28fb..e6b8041248a02ca07e20c62074233b2f23b2d663 100644
--- a/mace/utils/command_line_flags.cc
+++ b/mace/utils/command_line_flags.cc
@@ -3,32 +3,32 @@
 //
 
 #include "mace/utils/command_line_flags.h"
-#include "mace/utils/logging.h"
 
 #include <cstring>
 #include <iomanip>
 
-namespace mace {
-namespace {
+#include "mace/utils/logging.h"
 
-using namespace std;
+namespace mace {
+namespace utils {
 
-bool StringConsume(string &arg, const string &x) {
-  if ((arg.size() >= x.size()) &&
-      (memcmp(arg.data(), x.data(), x.size()) == 0)) {
-    arg = arg.substr(x.size());
+bool StringConsume(const std::string &x, std::string *arg) {
+  MACE_CHECK_NOTNULL(arg);
+  if ((arg->size() >= x.size()) &&
+      (memcmp(arg->data(), x.data(), x.size()) == 0)) {
+    *arg = arg->substr(x.size());
     return true;
   }
   return false;
 }
 
-bool ParseStringFlag(string arg,
-                     string flag,
-                     string *dst,
+bool ParseStringFlag(std::string arg,
+                     std::string flag,
+                     std::string *dst,
                      bool *value_parsing_ok) {
   *value_parsing_ok = true;
-  if (StringConsume(arg, "--") && StringConsume(arg, flag) &&
-      StringConsume(arg, "=")) {
+  if (StringConsume("--", &arg) && StringConsume(flag, &arg) &&
+      StringConsume("=", &arg)) {
     *dst = arg;
     return true;
   }
@@ -36,13 +36,13 @@ bool ParseStringFlag(string arg,
   return false;
 }
 
-bool ParseInt32Flag(string arg,
-                    string flag,
+bool ParseInt32Flag(std::string arg,
+                    std::string flag,
                     int32_t *dst,
                     bool *value_parsing_ok) {
   *value_parsing_ok = true;
-  if (StringConsume(arg, "--") && StringConsume(arg, flag) &&
-      StringConsume(arg, "=")) {
+  if (StringConsume("--", &arg) && StringConsume(flag, &arg) &&
+      StringConsume("=", &arg)) {
     char extra;
     if (sscanf(arg.data(), "%d%c", dst, &extra) != 1) {
       LOG(ERROR) << "Couldn't interpret value " << arg << " for flag " << flag
@@ -55,13 +55,13 @@ bool ParseInt32Flag(string arg,
   return false;
 }
 
-bool ParseInt64Flag(string arg,
-                    string flag,
-                    long long *dst,
+bool ParseInt64Flag(std::string arg,
+                    std::string flag,
+                    int64_t *dst,
                     bool *value_parsing_ok) {
   *value_parsing_ok = true;
-  if (StringConsume(arg, "--") && StringConsume(arg, flag) &&
-      StringConsume(arg, "=")) {
+  if (StringConsume("--", &arg) && StringConsume(flag, &arg) &&
+      StringConsume("=", &arg)) {
     char extra;
     if (sscanf(arg.data(), "%lld%c", dst, &extra) != 1) {
       LOG(ERROR) << "Couldn't interpret value " << arg << " for flag " << flag
@@ -74,9 +74,10 @@ bool ParseInt64Flag(string arg,
   return false;
 }
 
-bool ParseBoolFlag(string arg, string flag, bool *dst, bool *value_parsing_ok) {
+bool ParseBoolFlag(std::string arg, std::string flag,
+                   bool *dst, bool *value_parsing_ok) {
   *value_parsing_ok = true;
-  if (StringConsume(arg, "--") && StringConsume(arg, flag)) {
+  if (StringConsume("--", &arg) && StringConsume(flag, &arg)) {
     if (arg.empty()) {
       *dst = true;
       return true;
@@ -99,13 +100,13 @@ bool ParseBoolFlag(string arg, string flag, bool *dst, bool *value_parsing_ok) {
   return false;
 }
 
-bool ParseFloatFlag(string arg,
-                    string flag,
+bool ParseFloatFlag(std::string arg,
+                    std::string flag,
                     float *dst,
                     bool *value_parsing_ok) {
   *value_parsing_ok = true;
-  if (StringConsume(arg, "--") && StringConsume(arg, flag) &&
-      StringConsume(arg, "=")) {
+  if (StringConsume("--", &arg) && StringConsume(flag, &arg) &&
+      StringConsume("=", &arg)) {
     char extra;
     if (sscanf(arg.data(), "%f%c", dst, &extra) != 1) {
       LOG(ERROR) << "Couldn't interpret value " << arg << " for flag " << flag
@@ -118,47 +119,48 @@ bool ParseFloatFlag(string arg,
   return false;
 }
 
-}  // namespace
+}  // namespace utils
 
-Flag::Flag(const char *name, int *dst, const string &usage_text)
+Flag::Flag(const char *name, int *dst, const std::string &usage_text)
     : name_(name), type_(TYPE_INT), int_value_(dst), usage_text_(usage_text) {}
 
-Flag::Flag(const char *name, long long *dst, const string &usage_text)
+Flag::Flag(const char *name, int64_t *dst, const std::string &usage_text)
     : name_(name),
       type_(TYPE_INT64),
       int64_value_(dst),
       usage_text_(usage_text) {}
 
-Flag::Flag(const char *name, bool *dst, const string &usage_text)
+Flag::Flag(const char *name, bool *dst, const std::string &usage_text)
     : name_(name),
       type_(TYPE_BOOL),
       bool_value_(dst),
       usage_text_(usage_text) {}
 
-Flag::Flag(const char *name, string *dst, const string &usage_text)
+Flag::Flag(const char *name, std::string *dst, const std::string &usage_text)
     : name_(name),
       type_(TYPE_STRING),
       string_value_(dst),
       usage_text_(usage_text) {}
 
-Flag::Flag(const char *name, float *dst, const string &usage_text)
+Flag::Flag(const char *name, float *dst, const std::string &usage_text)
     : name_(name),
       type_(TYPE_FLOAT),
       float_value_(dst),
       usage_text_(usage_text) {}
 
-bool Flag::Parse(string arg, bool *value_parsing_ok) const {
+bool Flag::Parse(std::string arg, bool *value_parsing_ok) const {
   bool result = false;
   if (type_ == TYPE_INT) {
-    result = ParseInt32Flag(arg, name_, int_value_, value_parsing_ok);
+    result = utils::ParseInt32Flag(arg, name_, int_value_, value_parsing_ok);
   } else if (type_ == TYPE_INT64) {
-    result = ParseInt64Flag(arg, name_, int64_value_, value_parsing_ok);
+    result = utils::ParseInt64Flag(arg, name_, int64_value_, value_parsing_ok);
   } else if (type_ == TYPE_BOOL) {
-    result = ParseBoolFlag(arg, name_, bool_value_, value_parsing_ok);
+    result = utils::ParseBoolFlag(arg, name_, bool_value_, value_parsing_ok);
   } else if (type_ == TYPE_STRING) {
-    result = ParseStringFlag(arg, name_, string_value_, value_parsing_ok);
+    result = utils::ParseStringFlag(arg, name_,
+                                    string_value_, value_parsing_ok);
   } else if (type_ == TYPE_FLOAT) {
-    result = ParseFloatFlag(arg, name_, float_value_, value_parsing_ok);
+    result = utils::ParseFloatFlag(arg, name_, float_value_, value_parsing_ok);
   }
   return result;
 }
@@ -169,7 +171,7 @@ bool Flag::Parse(string arg, bool *value_parsing_ok) const {
   bool result = true;
   std::vector<char *> unknown_flags;
   for (int i = 1; i < *argc; ++i) {
-    if (string(argv[i]) == "--") {
+    if (std::string(argv[i]) == "--") {
       while (i < *argc) {
         unknown_flags.push_back(argv[i]);
         ++i;
@@ -202,8 +204,8 @@ bool Flag::Parse(string arg, bool *value_parsing_ok) const {
   return result && (*argc < 2 || strcmp(argv[1], "--help") != 0);
 }
 
-/*static*/ string Flags::Usage(const string &cmdline,
-                               const std::vector<Flag> &flag_list) {
+std::string Flags::Usage(const std::string &cmdline,
+                         const std::vector<Flag> &flag_list) {
   std::stringstream usage_text;
   usage_text << "usage: " << cmdline << std::endl;
 
diff --git a/mace/utils/command_line_flags.h b/mace/utils/command_line_flags.h
index 4373ceedef7e58ad72ec86290dad103d5e6df992..3fa5a9e83cad1814c4dcb7eca4099e7f98a86a1d 100644
--- a/mace/utils/command_line_flags.h
+++ b/mace/utils/command_line_flags.h
@@ -2,8 +2,8 @@
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
 
-#ifndef MACE_CORE_COMMAND_LINE_FLAGS_H
-#define MACE_CORE_COMMAND_LINE_FLAGS_H
+#ifndef MACE_UTILS_COMMAND_LINE_FLAGS_H_
+#define MACE_UTILS_COMMAND_LINE_FLAGS_H_
 
 #include <string>
 #include <vector>
@@ -13,7 +13,7 @@ namespace mace {
 class Flag {
  public:
   Flag(const char *name, int *dst1, const std::string &usage_text);
-  Flag(const char *name, long long *dst1, const std::string &usage_text);
+  Flag(const char *name, int64_t *dst1, const std::string &usage_text);
   Flag(const char *name, bool *dst, const std::string &usage_text);
   Flag(const char *name, std::string *dst, const std::string &usage_text);
   Flag(const char *name, float *dst, const std::string &usage_text);
@@ -26,7 +26,7 @@ class Flag {
   std::string name_;
   enum { TYPE_INT, TYPE_INT64, TYPE_BOOL, TYPE_STRING, TYPE_FLOAT } type_;
   int *int_value_;
-  long long *int64_value_;
+  int64_t *int64_value_;
   bool *bool_value_;
   std::string *string_value_;
   float *float_value_;
@@ -50,4 +50,4 @@ class Flags {
 
 }  // namespace mace
 
-#endif  // MACE_CORE_COMMAND_LINE_FLAGS_H
+#endif  // MACE_UTILS_COMMAND_LINE_FLAGS_H_
diff --git a/mace/utils/env_time.h b/mace/utils/env_time.h
index ce70a244d3eb2942ead7cf8285d86845ce1d4e91..0fecab568e4a48ca138802e65d59142538946167 100644
--- a/mace/utils/env_time.h
+++ b/mace/utils/env_time.h
@@ -2,8 +2,8 @@
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
 
-#ifndef MACE_UTILS_ENV_TIME_H
-#define MACE_UTILS_ENV_TIME_H
+#ifndef MACE_UTILS_ENV_TIME_H_
+#define MACE_UTILS_ENV_TIME_H_
 
 #include <stdint.h>
 #include <sys/time.h>
@@ -19,4 +19,4 @@ inline int64_t NowMicros() {
 
 }  // namespace mace
 
-#endif  // MACE_UTILS_ENV_TIME_H
+#endif  // MACE_UTILS_ENV_TIME_H_
diff --git a/mace/utils/logging.h b/mace/utils/logging.h
index 22e39488b7bdd1a83a9d1c89cc6c74345bd751fd..4a6da3befb985a57d702a1766799722f6271bb36 100644
--- a/mace/utils/logging.h
+++ b/mace/utils/logging.h
@@ -9,6 +9,7 @@
 #include <sstream>
 #include <string>
 #include <vector>
+#include <utility>
 
 #include "mace/public/mace.h"
 #include "mace/utils/env_time.h"
@@ -122,7 +123,7 @@ class LatencyLogger {
       VLOG(vlog_level_) << message_
                         << " latency: " << stop_micros - start_micros_ << " us";
     }
-  };
+  }
 
  private:
   const int vlog_level_;
diff --git a/mace/utils/memory_logging.h b/mace/utils/memory_logging.h
index e9d540bccc9ab14607dc39ca55d0ed4ca160d50f..bd46f58049d13ed6b7cbda455ad7c883a7e33f88 100644
--- a/mace/utils/memory_logging.h
+++ b/mace/utils/memory_logging.h
@@ -6,13 +6,15 @@
 #define MACE_UTILS_MEMORY_LOGGING_H_
 
 #include <malloc.h>
+#include <string>
+
 #include "mace/utils/logging.h"
 
 namespace mace {
 
 class MallinfoChangeLogger {
  public:
-  MallinfoChangeLogger(const std::string &name) : name_(name) {
+  explicit MallinfoChangeLogger(const std::string &name) : name_(name) {
     prev_ = mallinfo();
   }
   ~MallinfoChangeLogger() {
diff --git a/mace/utils/string_util.h b/mace/utils/string_util.h
index aad884d3349f35d261a0292e4686baa9b16379a0..ee1b5918383e7c494c14299c95ea5cf8e6af132f 100644
--- a/mace/utils/string_util.h
+++ b/mace/utils/string_util.h
@@ -10,7 +10,7 @@
 #include <vector>
 
 namespace mace {
-namespace {
+namespace string_util {
 
 inline void MakeStringInternal(std::stringstream & /*ss*/) {}
 
@@ -27,12 +27,12 @@ inline void MakeStringInternal(std::stringstream &ss,
   MakeStringInternal(ss, args...);
 }
 
-}  // namespace
+}  // namespace string_util
 
 template <typename... Args>
 std::string MakeString(const Args &... args) {
   std::stringstream ss;
-  MakeStringInternal(ss, args...);
+  string_util::MakeStringInternal(ss, args...);
   return ss.str();
 }
 
diff --git a/mace/utils/tuner.h b/mace/utils/tuner.h
index fee8cd95a27dd27c3773d133c2c098bc8a2257e4..1b05f3a3199b04def2a8d17f255d0bb82a7acbae 100644
--- a/mace/utils/tuner.h
+++ b/mace/utils/tuner.h
@@ -17,8 +17,6 @@
 #include "mace/utils/timer.h"
 #include "mace/utils/utils.h"
 
-namespace {}  // namespace
-
 namespace mace {
 
 extern bool GetTuningParams(
@@ -127,7 +125,7 @@ class Tuner {
       const std::function<RetType(const std::vector<param_type> &,
                                   Timer *,
                                   std::vector<param_type> *)> &func,
-      std::vector<param_type> &params,
+      const std::vector<param_type> &params,
       Timer *timer,
       int num_runs,
       double *time_us,
diff --git a/mace/utils/tuner_development.cc b/mace/utils/tuner_development.cc
index ccbe3b64fc5249e1de6e5885b169bb2af8f5be67..570b94cd465af7980a62d0e9eb92e10172c427fa 100644
--- a/mace/utils/tuner_development.cc
+++ b/mace/utils/tuner_development.cc
@@ -1,3 +1,6 @@
+//
+// Copyright (c) 2017 XiaoMi All rights reserved.
+//
 #include <fstream>
 #include <map>
 #include <string>
diff --git a/mace/utils/tuner_production.cc b/mace/utils/tuner_production.cc
index f008d8dc52213d644f32d31a6efb68e37deef3a7..c45c6d7e1edd0d7a959bf63e61897b4e04aeb167 100644
--- a/mace/utils/tuner_production.cc
+++ b/mace/utils/tuner_production.cc
@@ -1,3 +1,6 @@
+//
+// Copyright (c) 2017 XiaoMi All rights reserved.
+//
 #include <map>
 #include <string>
 #include <unordered_map>
diff --git a/mace/utils/tuner_test.cc b/mace/utils/tuner_test.cc
index 80cc50d08ab57b8601207d516afb31f3ca10a722..b040a5a62694e53f0792088080d3a7c644c4dd7d 100644
--- a/mace/utils/tuner_test.cc
+++ b/mace/utils/tuner_test.cc
@@ -1,7 +1,7 @@
 //
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
-#include <thread>
+#include <thread>  // NOLINT(build/c++11)
 
 #include "gtest/gtest.h"
 
diff --git a/mace/utils/utils.h b/mace/utils/utils.h
index 0330de47a052569a88855b3a976f61055e2c57ea..54ab5793d229bba9453ae38fae6ec260f3c9cb6e 100644
--- a/mace/utils/utils.h
+++ b/mace/utils/utils.h
@@ -7,6 +7,8 @@
 
 #include <sys/time.h>
 #include <sstream>
+#include <string>
+#include <utility>
 
 namespace mace {
 template <typename Integer>
diff --git a/mace/utils/utils_test.cc b/mace/utils/utils_test.cc
index 80cc50d08ab57b8601207d516afb31f3ca10a722..b040a5a62694e53f0792088080d3a7c644c4dd7d 100644
--- a/mace/utils/utils_test.cc
+++ b/mace/utils/utils_test.cc
@@ -1,7 +1,7 @@
 //
 // Copyright (c) 2017 XiaoMi All rights reserved.
 //
-#include <thread>
+#include <thread>  // NOLINT(build/c++11)
 
 #include "gtest/gtest.h"