Add bigger models to performance benchmarks (#342)

* add larger test models

test/backend_performance.cpp

@@ -27,6 +27,7 @@
 #include "ngraph/runtime/call_frame.hpp"
 #include "ngraph/runtime/cpu/cpu_call_frame.hpp"
 #include "ngraph/runtime/manager.hpp"
+#include "ngraph/runtime/utils.hpp"
 #include "ngraph/serializer.hpp"
 #include "ngraph/util.hpp"
 #include "util/random.hpp"
@@ -59,51 +60,32 @@ static multimap<size_t, string>
     return rc;
 }
 
-// Starting point CPU: 1.2ms/iteration
-
-shared_ptr<runtime::TensorView> make_tensor(runtime::Backend& backend, const ValueType& value)
-{
-    shared_ptr<runtime::TensorView> arg =
-        backend.make_primary_tensor_view(value.get_element_type(), value.get_shape());
-    return arg;
-}
-
-TEST(benchmark, mxnet_mnist_mlp_forward)
+shared_ptr<runtime::Value> make_value(runtime::Backend& backend, const ValueType& value)
 {
-    test::Uniform<float> rng{-1, 1, 0};
-    const string json_path = file_util::path_join(SERIALIZED_ZOO, "mxnet/mnist_mlp_forward.json");
-    const string json_string = file_util::read_file_to_string(json_path);
-    stringstream ss(json_string);
-    shared_ptr<Function> f = ngraph::deserialize(ss);
-
-    auto manager = runtime::Manager::get("CPU");
-    auto external = manager->compile(f);
-    auto backend = manager->allocate_backend();
-    auto cf = backend->make_call_frame(external);
-
-    vector<shared_ptr<runtime::Value>> args;
-    for (shared_ptr<op::Parameter> param : f->get_parameters())
+    shared_ptr<runtime::Value> rc;
+    const TupleType* tuple = dynamic_cast<const TupleType*>(&value);
+    const TensorViewType* tensor = dynamic_cast<const TensorViewType*>(&value);
+    if (tuple)
     {
-        auto arg = make_tensor(*backend, *(param->get_value_type()));
-        rng.initialize(arg);
-        args.push_back(arg);
+        vector<shared_ptr<runtime::Value>> elements;
+        for (auto value_type : tuple->get_element_types())
+        {
+            elements.push_back(make_value(backend, *value_type));
         }
-    shared_ptr<const ValueType> result_type = f->get_result_type();
-    auto result = make_tensor(*backend, *result_type);
-
-    stopwatch t1;
-    t1.start();
-    float count = 1000;
-    for (size_t i = 0; i < static_cast<size_t>(count); i++)
+        rc = backend.make_tuple(elements);
+    }
+    else if (tensor)
     {
-        cf->call(args, {result});
+        rc = backend.make_primary_tensor_view(value.get_element_type(), value.get_shape());
     }
-    t1.stop();
-    float time = t1.get_milliseconds();
-    cout << time / count << "ms per iteration\n";
+    else
+    {
+        NGRAPH_INFO;
+    }
+    return rc;
 }
 
-TEST(benchmark, mxnet_10_bucket_lstm)
+void run_benchmark(const std::string& json_path, size_t iterations)
 {
     bool emit_timing = (std::getenv("NGRAPH_CPU_EMIT_TIMING") != nullptr);
     if (!emit_timing)
@@ -112,37 +94,52 @@ TEST(benchmark, mxnet_10_bucket_lstm)
     }
 
     test::Uniform<float> rng{-1, 1, 0};
-    const string json_path = file_util::path_join(SERIALIZED_ZOO, "mxnet/10_bucket_LSTM.json");
     const string json_string = file_util::read_file_to_string(json_path);
     stringstream ss(json_string);
     shared_ptr<Function> f = ngraph::deserialize(ss);
 
+    stopwatch build_time;
+    build_time.start();
     auto manager = runtime::Manager::get("CPU");
     auto external = manager->compile(f);
     auto backend = manager->allocate_backend();
     auto cf = backend->make_call_frame(external);
     runtime::cpu::CPU_CallFrame* cpu_cf = static_cast<runtime::cpu::CPU_CallFrame*>(cf.get());
+    build_time.stop();
+    cout << "build_time " << build_time.get_milliseconds() << "ms" << endl;
 
     vector<shared_ptr<runtime::Value>> args;
     for (shared_ptr<op::Parameter> param : f->get_parameters())
     {
-        auto arg = make_tensor(*backend, *(param->get_value_type()));
-        rng.initialize(arg);
+        auto arg = make_value(*backend, *(param->get_value_type()));
+        auto tuple = dynamic_pointer_cast<runtime::Tuple>(arg);
+        auto tensor = dynamic_pointer_cast<runtime::TensorView>(arg);
+        if (tuple)
+        {
+            NGRAPH_INFO;
+            // Need to initialize this...maybe later
+        }
+        else if (tensor)
+        {
+            rng.initialize(tensor);
+        }
+        else
+        {
+        }
         args.push_back(arg);
     }
     shared_ptr<const ValueType> result_type = f->get_result_type();
-    auto result = make_tensor(*backend, *result_type);
+    auto result = make_value(*backend, *result_type);
 
     stopwatch t1;
     t1.start();
-    float count = 30;
-    for (size_t i = 0; i < static_cast<size_t>(count); i++)
+    for (size_t i = 0; i < static_cast<size_t>(iterations); i++)
     {
         cf->call(args, {result});
     }
     t1.stop();
     float time = t1.get_milliseconds();
-    cout << time / count << "ms per iteration\n";
+    cout << time / iterations << "ms per iteration" << endl;
 
     if (emit_timing)
     {
@@ -160,20 +157,33 @@ TEST(benchmark, mxnet_10_bucket_lstm)
             cout << setw(15) << left << it->second << " " << setw(10) << right << it->first
                  << "us\n";
         }
-        // size_t count = 30;
-        // for (const runtime::cpu::PerformanceCounter& p : perf_data)
-        // {
-        //     if (--count == 0)
-        //     {
-        //         break;
-        //     }
-        //     cout.imbue(locale(""));
-        //     cout << setw(15) << left << p.name() << " " << setw(10) << right
-        //          << p.microseconds() << "us\n";
-        // }
     }
 }
 
+TEST(benchmark, mxnet_mnist_mlp_forward)
+{
+    const string json_path = file_util::path_join(SERIALIZED_ZOO, "mxnet/mnist_mlp_forward.json");
+    run_benchmark(json_path, 1000);
+}
+
+TEST(benchmark, mxnet_10_bucket_lstm)
+{
+    const string json_path = file_util::path_join(SERIALIZED_ZOO, "mxnet/10_bucket_LSTM.json");
+    run_benchmark(json_path, 10);
+}
+
+TEST(benchmark, mxnet_lstm_backward)
+{
+    const string json_path = file_util::path_join(SERIALIZED_ZOO, "mxnet/LSTM_backward.json");
+    run_benchmark(json_path, 10);
+}
+
+TEST(benchmark, mxnet_lstm_forward)
+{
+    const string json_path = file_util::path_join(SERIALIZED_ZOO, "mxnet/LSTM_forward.json");
+    run_benchmark(json_path, 10);
+}
+
 //
 // Benchmarks a graph that concatenates six 32x1x200 arrays along the middle axis.
 //

test/eigen.cpp

@@ -31,20 +31,3 @@ TEST(eigen, simple)
     m(1, 1) = m(1, 0) + m(0, 1);
     EXPECT_FLOAT_EQ(m(1, 1), 1.5);
 }
-
-TEST(eigen, test)
-{
-    float arg0[] = {1, 2, 3, 4, 5, 6, 7, 8};
-    float arg1[] = {1, 2, 3, 4, 5, 6, 7, 8};
-    vector<float> out(8);
-
-    Eigen::Map<Eigen::Array<float, 8, 1>, Eigen::Unaligned>(out.data()) =
-        Eigen::Map<Eigen::Array<float, 8, 1>, Eigen::Unaligned>(arg0) *
-        Eigen::Map<Eigen::Array<float, 8, 1>, Eigen::Unaligned>(arg1);
-
-    cout << "result " << join(out) << endl;
-
-    // EigenArray1d<float>(Multiply_163_0, fmt::V{12}) =
-    //     EigenArray1d<float>(Maximum_162_0, fmt::V{12}) *
-    //     EigenArray1d<float>(Parameter_158_0, fmt::V{12});
-}

test/serialize.cpp

@@ -126,7 +126,10 @@ TEST(serialize, main)
 
 TEST(serialize, existing_models)
 {
-    vector<string> models = {"mxnet/mnist_mlp_forward.json", "mxnet/10_bucket_LSTM.json"};
+    vector<string> models = {"mxnet/mnist_mlp_forward.json",
+                             "mxnet/10_bucket_LSTM.json",
+                             "mxnet/LSTM_backward.json",
+                             "mxnet/LSTM_forward.json"};
 
     for (const string& model : models)
     {