Jmenon/nbench (#1365)

* Add an option to exclude the first iteration

* Switch to warmup iterations

* Cleanup

src/tools/nbench/benchmark.cpp

@@ -83,7 +83,8 @@ multimap<size_t, string> aggregate_timing(const vector<runtime::PerformanceCount
 void run_benchmark(const string& json_path,
                    const string& backend_name,
                    size_t iterations,
-                   bool timing_detail)
+                   bool timing_detail,
+                   int warmup_iterations)
 {
     stopwatch timer;
     timer.start();
@@ -92,7 +93,7 @@ void run_benchmark(const string& json_path,
     shared_ptr<Function> f = deserialize(ss);
     timer.stop();
     cout << "deserialize time: " << timer.get_milliseconds() << "ms" << endl;
-    run_benchmark(f, backend_name, iterations, timing_detail);
+    run_benchmark(f, backend_name, iterations, timing_detail, warmup_iterations);
 }
 
 void print_times(const multimap<size_t, string>& timing)
@@ -238,7 +239,8 @@ static void random_init(shared_ptr<runtime::TensorView> tv)
 void run_benchmark(shared_ptr<Function> f,
                    const string& backend_name,
                    size_t iterations,
-                   bool timing_detail)
+                   bool timing_detail,
+                   int warmup_iterations)
 {
     stopwatch timer;
     timer.start();
@@ -272,6 +274,15 @@ void run_benchmark(shared_ptr<Function> f,
             args[i]->set_stale(false);
         }
     }
+
+    if (warmup_iterations)
+    {
+        for (int i = 0; i < warmup_iterations; i++)
+        {
+            backend->call(f, results, args);
+        }
+    }
+
     stopwatch t1;
     t1.start();
     for (size_t i = 0; i < static_cast<size_t>(iterations); i++)

src/tools/nbench/benchmark.hpp

@@ -31,9 +31,11 @@ std::multimap<size_t, std::string>
 void run_benchmark(std::shared_ptr<ngraph::Function> f,
                    const std::string& backend_name,
                    size_t iterations,
-                   bool timing_detail);
+                   bool timing_detail,
+                   int warmup_iterations);
 
 void run_benchmark(const std::string& json_path,
                    const std::string& backend_name,
                    size_t iterations,
-                   bool timing_detail = false);
+                   bool timing_detail = false,
+                   int warmup_iterations = 1);

src/tools/nbench/nbench.cpp

@@ -43,6 +43,8 @@ int main(int argc, char** argv)
     bool statistics = false;
     bool timing_detail = false;
     bool visualize = false;
+    int warmup_iterations = 1;
+
     for (size_t i = 1; i < argc; i++)
     {
         string arg = argv[i];
@@ -82,6 +84,18 @@ int main(int argc, char** argv)
         {
             directory = argv[++i];
         }
+        else if (arg == "-w" || arg == "--warmup_iterations")
+        {
+            try
+            {
+                warmup_iterations = stoi(argv[++i]);
+            }
+            catch (...)
+            {
+                cout << "Invalid Argument\n";
+                failed = true;
+            }
+        }
         else
         {
             cout << "Unknown option: " << arg << endl;
@@ -114,13 +128,14 @@ SYNOPSIS
         nbench [-f <filename>] [-b <backend>] [-i <iterations>]
 
 OPTIONS
-        -f|--file          Serialized model file
-        -b|--backend       Backend to use (default: CPU)
-        -d|--directory     Directory to scan for models. All models are benchmarked.
-        -i|--iterations    Iterations (default: 10)
-        -s|--statistics    Display op stastics
-        -v|--visualize     Visualize a model (WARNING: requires GraphViz installed)
-        --timing-detail    Gather detailed timing
+        -f|--file                 Serialized model file
+        -b|--backend              Backend to use (default: CPU)
+        -d|--directory            Directory to scan for models. All models are benchmarked.
+        -i|--iterations           Iterations (default: 10)
+        -s|--statistics           Display op stastics
+        -v|--visualize            Visualize a model (WARNING: requires GraphViz installed)
+        --timing-detail           Gather detailed timing
+        -w|--warmup_iterations    Number of warm-up iterations
 )###";
         return 1;
     }
@@ -190,7 +205,7 @@ OPTIONS
                 shared_ptr<Function> f = deserialize(m);
                 cout << "Benchmarking " << m << ", " << backend << " backend, " << iterations
                      << " iterations.\n";
-                run_benchmark(f, backend, iterations, timing_detail);
+                run_benchmark(f, backend, iterations, timing_detail, warmup_iterations);
             }
             catch (exception e)
             {
@@ -203,7 +218,7 @@ OPTIONS
         shared_ptr<Function> f = deserialize(model);
         cout << "Benchmarking " << model << ", " << backend << " backend, " << iterations
              << " iterations.\n";
-        run_benchmark(f, backend, iterations, timing_detail);
+        run_benchmark(f, backend, iterations, timing_detail, warmup_iterations);
     }
 
     return 0;