fix(lite): fix ldr use lite interface error when open both fast-run and nchw44

GitOrigin-RevId: 27b29d60af

lite/load_and_run/src/helpers/common.h

@@ -11,27 +11,29 @@ enum class RunStage {
 
     BEFORE_MODEL_LOAD = 0,
 
-    AFTER_MODEL_LOAD = 1,
+    AFTER_NETWORK_CREATED = 1,
 
-    BEFORE_OUTSPEC_SET = 2,
+    AFTER_MODEL_LOAD = 2,
+
+    BEFORE_OUTSPEC_SET = 3,
 
     //! using for dump static memory information svg file
-    AFTER_OUTSPEC_SET = 3,
+    AFTER_OUTSPEC_SET = 4,
 
     //! using for external c opr library
-    MODEL_RUNNING = 4,
+    MODEL_RUNNING = 5,
 
     //! using for output dumper
-    AFTER_RUNNING_WAIT = 5,
+    AFTER_RUNNING_WAIT = 6,
 
     //! using for external c opr library
-    AFTER_RUNNING_ITER = 6,
+    AFTER_RUNNING_ITER = 7,
 
-    AFTER_MODEL_RUNNING = 7,
+    AFTER_MODEL_RUNNING = 8,
 
-    GLOBAL_OPTIMIZATION = 8,
+    GLOBAL_OPTIMIZATION = 9,
 
-    UPDATE_IO = 9,
+    UPDATE_IO = 10,
 };
 /*!
  * \brief: type of different model

lite/load_and_run/src/models/model.h

@@ -24,6 +24,8 @@ public:
 
     virtual void set_shared_mem(bool state) = 0;
 
+    virtual void create_network(){};
+
     //! load model interface for load and run strategy
     virtual void load_model() = 0;
 

lite/load_and_run/src/models/model_lite.cpp

@@ -10,12 +10,12 @@ using namespace lar;
 ModelLite::ModelLite(const std::string& path) : model_path(path) {
     LITE_LOG("creat lite model use CPU as default comp node");
 };
-void ModelLite::load_model() {
+
+void ModelLite::create_network() {
     m_network = std::make_shared<lite::Network>(config, IO);
-    if (enable_layout_transform) {
-        LITE_LOG("enable layout transform while load model for lite");
-        lite::Runtime::enable_global_layout_transform(m_network);
-    }
+}
+
+void ModelLite::load_model() {
     if (share_model_mem) {
         //! WARNNING:maybe not right to share param memmory for this
         LITE_LOG("enable share model memory");
@@ -116,4 +116,4 @@ std::vector<uint8_t> ModelLite::get_model_data() {
     LITE_THROW("unsupported interface: ModelLite::get_model_data() \n");
 
     return out_data;
-}
+}

lite/load_and_run/src/models/model_lite.h

@@ -22,6 +22,9 @@ public:
     void set_shared_mem(bool state) override { share_model_mem = state; }
 
     //! load model from dump file
+    void create_network() override;
+
+    //! load model from dump file
     void load_model() override;
 
     //! run model with given runtime parameter
@@ -34,9 +37,6 @@ public:
     std::shared_ptr<mgb::json::Object> get_io_info() override;
 #endif
 
-    //! enable global layout transform
-    void set_layout_transform(bool state) { enable_layout_transform = state; }
-
     //! get the network of lite model
     std::shared_ptr<lite::Network>& get_lite_network() { return m_network; }
 
@@ -61,7 +61,6 @@ public:
 
 private:
     bool share_model_mem = false;
-    bool enable_layout_transform = false;
     std::string model_path;
 
     DataParser parser;

lite/load_and_run/src/options/fastrun_options.cpp

@@ -19,7 +19,7 @@ namespace lar {
 template <>
 void FastRunOption::config_model_internel<ModelLite>(
         RuntimeParam& runtime_param, std::shared_ptr<ModelLite> model) {
-    if (runtime_param.stage == RunStage::BEFORE_MODEL_LOAD) {
+    if (runtime_param.stage == RunStage::AFTER_NETWORK_CREATED) {
         //! set the algo policy before model load
         using Strategy = ModelLite::Strategy;
         uint32_t strategy = 0;
@@ -44,23 +44,17 @@ void FastRunOption::config_model_internel<ModelLite>(
                            strategy;
         }
         auto lite_strategy = static_cast<Strategy>(strategy);
-        model->set_lite_strategy(lite_strategy);
-    } else if (runtime_param.stage == RunStage::AFTER_MODEL_LOAD) {
-        auto&& lite_network = model->get_lite_network();
-        auto&& lite_strategy = model->get_lite_strategy();
         //! set algo policy for model
+        auto&& lite_network = model->get_lite_network();
         lite::Runtime::set_network_algo_policy(
                 lite_network, lite_strategy, share_batch_size, batch_binary_equal);
+    } else if (runtime_param.stage == RunStage::AFTER_MODEL_LOAD) {
         if (!m_fast_run_cache.empty()) {
             if (!access(m_fast_run_cache.c_str(), F_OK)) {
                 lite::set_persistent_cache(m_fast_run_cache);
             } else {
                 lite::set_persistent_cache(m_fast_run_cache, true);
             }
-            //! TODO:this is from mdl model settings but not matched settings in
-            //! lite model
-            // if (!enable_full_run && !enable_fast_run)
-            //     mgb::gopt::enable_opr_use_profiling_cache_inplace(vars);
         }
     } else if (runtime_param.stage == RunStage::AFTER_MODEL_RUNNING) {
 #if MGB_ENABLE_FASTRUN
@@ -255,4 +249,4 @@ DEFINE_int32(fast_run_shared_batch_size, 0, "Set the batch size used during fast
 DEFINE_string(fast_run_algo_policy, "", "fast-run cache path.");
 
 REGIST_OPTION_CREATOR(fastrun, lar::FastRunOption::create_option);
-REGIST_OPTION_VALIDATER(fastrun, lar::FastRunOption::set_valid);
+REGIST_OPTION_VALIDATER(fastrun, lar::FastRunOption::set_valid);

lite/load_and_run/src/options/layout_trans_options.cpp

@@ -9,7 +9,7 @@ namespace lar {
 template <>
 void GoptLayoutOption::config_model_internel<ModelLite>(
         RuntimeParam& runtime_param, std::shared_ptr<ModelLite> model) {
-    if (runtime_param.stage == RunStage::BEFORE_MODEL_LOAD) {
+    if (runtime_param.stage == RunStage::AFTER_NETWORK_CREATED) {
         if (m_layout_transform) {
             LITE_LOG("using global layout transform optimization\n");
             if (m_layout_transform_target ==
@@ -23,7 +23,9 @@ void GoptLayoutOption::config_model_internel<ModelLite>(
                 model->get_config().device_type = LiteDeviceType::LITE_CUDA;
             }
 #endif
-            model->set_layout_transform(true);
+            LITE_LOG("enable layout transform while load model for lite");
+            auto&& lite_network = model->get_lite_network();
+            lite::Runtime::enable_global_layout_transform(lite_network);
         }
     } else if (runtime_param.stage == RunStage::GLOBAL_OPTIMIZATION) {
         if (m_layout_transform) {
@@ -266,4 +268,4 @@ DEFINE_int32(
         layout_transform_batch_size, -1,
         "the batch size of input for global layout transform optimization working on");
 REGIST_OPTION_CREATOR(gopt_layout, lar::GoptLayoutOption::create_option);
-REGIST_OPTION_VALIDATER(gopt_layout, lar::GoptLayoutOption::set_valid);
+REGIST_OPTION_VALIDATER(gopt_layout, lar::GoptLayoutOption::set_valid);

lite/load_and_run/src/strategys/strategy_fitting.cpp

@@ -197,6 +197,10 @@ void OptionsTimeProfiler::profile_with_given_options(
         runtime_param.stage = RunStage::BEFORE_MODEL_LOAD;
         stage_config_model();
 
+        runtime_param.stage = RunStage::AFTER_NETWORK_CREATED;
+        model->create_network();
+        stage_config_model();
+
         model->load_model();
         //! after load configure
         auto config_model_before_runing = [&]() {

lite/load_and_run/src/strategys/strategy_normal.cpp

@@ -42,6 +42,10 @@ void NormalStrategy::run_subline() {
     m_runtime_param.stage = RunStage::BEFORE_MODEL_LOAD;
     stage_config_model();
 
+    m_runtime_param.stage = RunStage::AFTER_NETWORK_CREATED;
+    model->create_network();
+    stage_config_model();
+
     mgb::RealTimer timer;
     model->load_model();
     mgb_log("load model: %.3fms\n", timer.get_msecs_reset());

lite/test/test_layout_options.cpp

@@ -18,6 +18,7 @@ DECLARE_bool(enable_nchw32);
 DECLARE_bool(enable_nchw64);
 DECLARE_bool(enable_nhwcd4);
 DECLARE_bool(enable_nchw44_dot);
+DECLARE_bool(fast_run);
 namespace {
 BOOL_OPTION_WRAP(enable_nchw4);
 BOOL_OPTION_WRAP(enable_chwn4);
@@ -27,6 +28,7 @@ BOOL_OPTION_WRAP(enable_nchw32);
 BOOL_OPTION_WRAP(enable_nchw64);
 BOOL_OPTION_WRAP(enable_nhwcd4);
 BOOL_OPTION_WRAP(enable_nchw44_dot);
+BOOL_OPTION_WRAP(fast_run);
 
 BOOL_OPTION_WRAP(lite);
 BOOL_OPTION_WRAP(cpu);
@@ -60,6 +62,17 @@ TEST(TestLarLayout, X86_CPU_LITE) {
     TEST_BOOL_OPTION(enable_nchw32);
     TEST_BOOL_OPTION(enable_nchw88);
 }
+
+TEST(TestLarLayoutFastRun, CPU_LITE) {
+    DEFINE_WRAP(cpu);
+    DEFINE_WRAP(lite);
+    std::string model_path = "./shufflenet.mge";
+    {
+        DEFINE_WRAP(enable_nchw44);
+        DEFINE_WRAP(fast_run);
+        run_NormalStrategy(model_path);
+    }
+}
 #if LITE_WITH_CUDA
 TEST(TestLarLayout, CUDA) {
     DEFINE_WRAP(cuda);

lite/test/test_options.h

@@ -25,9 +25,9 @@ void run_NormalStrategy(std::string model_path);
 
 #define DEFINE_WRAP(option) BoolOptionWrap_##option flags_##option;
 
-#define TEST_BOOL_OPTION(option)                \
-    {                                           \
-        BoolOptionWrap_##option flags_##option; \
-        run_NormalStrategy(model_path);         \
+#define TEST_BOOL_OPTION(option)        \
+    {                                   \
+        DEFINE_WRAP(option);            \
+        run_NormalStrategy(model_path); \
     }
 // vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}