Remove npu fast compile

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -632,7 +632,6 @@ typedef struct OrtMIGraphXProviderOptions {
 typedef struct OrtOpenVINOProviderOptions {
 #ifdef __cplusplus
   OrtOpenVINOProviderOptions() : device_type{},
-                                 enable_npu_fast_compile{},
                                  device_id{},
                                  num_of_threads{},
                                  cache_dir{},
@@ -645,7 +644,6 @@ typedef struct OrtOpenVINOProviderOptions {
    * Valid settings are one of: "CPU_FP32", "CPU_FP16", "GPU_FP32", "GPU_FP16"
    */
   const char* device_type;
-  unsigned char enable_npu_fast_compile;  ///< 0 = disabled, nonzero = enabled
   const char* device_id;
   size_t num_of_threads;  ///< 0 = Use default number of threads
   const char* cache_dir;  // path is set to empty by default

onnxruntime/core/providers/openvino/contexts.h

@@ -15,7 +15,6 @@ namespace openvino_ep {
 struct GlobalContext {
   OVCore ie_core;
   bool is_wholly_supported_graph = false;
-  bool enable_npu_fast_compile = false;
   bool enable_opencl_throttling = false;
   bool disable_dynamic_shapes = false;
   bool ep_context_embed_mode = true;

onnxruntime/core/providers/openvino/openvino_execution_provider.cc

@@ -25,7 +25,6 @@ OpenVINOExecutionProvider::OpenVINOExecutionProvider(const OpenVINOExecutionProv
   global_context_ = std::make_unique<openvino_ep::GlobalContext>();
   global_context_->device_type = info.device_type_;
   global_context_->precision_str = info.precision_;
-  global_context_->enable_npu_fast_compile = info.enable_npu_fast_compile_;
   global_context_->cache_dir = info.cache_dir_;
   global_context_->load_config = info.load_config_;
   global_context_->model_priority = info.model_priority_;

onnxruntime/core/providers/openvino/openvino_execution_provider.h

@@ -79,7 +79,6 @@ static std::vector<std::string> parseDevices(const std::string& device_string,
 struct OpenVINOExecutionProviderInfo {
   std::string device_type_{""};
   std::string precision_{""};
-  bool enable_npu_fast_compile_{false};
   size_t num_of_threads_{0};
   std::string load_config_{""};
   std::string cache_dir_{""};
@@ -96,15 +95,14 @@ struct OpenVINOExecutionProviderInfo {
   OpenVINOExecutionProviderInfo() = delete;
 
   explicit OpenVINOExecutionProviderInfo(const std::string& dev_type, const std::string& precision,
-                                         bool enable_npu_fast_compile, size_t num_of_threads,
+                                         size_t num_of_threads,
                                          const std::string& load_config, const std::string& cache_dir,
                                          const std::string& model_priority, int num_streams,
                                          void* context, bool enable_opencl_throttling,
                                          bool disable_dynamic_shapes, bool export_ep_ctx_blob,
                                          bool enable_qdq_optimizer, bool disable_cpu_fallback,
                                          bool so_epctx_embed_mode)
       : precision_(std::move(precision)),
-        enable_npu_fast_compile_(enable_npu_fast_compile),
         num_of_threads_(num_of_threads),
         load_config_(std::move(load_config)),
         cache_dir_(std::move(cache_dir)),

onnxruntime/core/providers/openvino/openvino_provider_factory.cc

@@ -9,14 +9,13 @@
 namespace onnxruntime {
 struct OpenVINOProviderFactory : IExecutionProviderFactory {
   OpenVINOProviderFactory(const std::string& device_type, const std::string& precision,
-                          bool enable_npu_fast_compile, size_t num_of_threads,
+                          size_t num_of_threads,
                           const std::string& load_config, const std::string& cache_dir,
                           const std::string& model_priority, int num_streams, void* context,
                           bool enable_opencl_throttling, bool disable_dynamic_shapes,
                           bool enable_qdq_optimizer, const ConfigOptions& config_options)
       : device_type_(device_type),
         precision_(precision),
-        enable_npu_fast_compile_(enable_npu_fast_compile),
         num_of_threads_(num_of_threads),
         load_config_(load_config),
         cache_dir_(cache_dir),
@@ -35,7 +34,6 @@ struct OpenVINOProviderFactory : IExecutionProviderFactory {
  private:
   std::string device_type_;
   std::string precision_;
-  bool enable_npu_fast_compile_;
   size_t num_of_threads_;
   std::string load_config_;
   std::string cache_dir_;
@@ -71,7 +69,7 @@ std::unique_ptr<IExecutionProvider> OpenVINOProviderFactory::CreateProvider() {
     }
   }
 
-  OpenVINOExecutionProviderInfo info(device_type_, precision_, enable_npu_fast_compile_, num_of_threads_, load_config_,
+  OpenVINOExecutionProviderInfo info(device_type_, precision_, num_of_threads_, load_config_,
                                      cache_dir_, model_priority_, num_streams_, context_, enable_opencl_throttling_,
                                      disable_dynamic_shapes_, so_export_ep_ctx_blob, enable_qdq_optimizer_,
                                      so_disable_cpu_fallback, so_epctx_embed_mode);
@@ -105,8 +103,6 @@ struct OpenVINO_Provider : Provider {
                                              // Not setting precision will execute with optimized precision for
                                              // best inference latency. set Precision=ACCURACY for executing models
                                              // with input precision for best accuracy.
-    bool enable_npu_fast_compile = false;    // [enable_npu_fast_compile]: Fast-compile may be optionally enabled to
-                                             // speeds up the model's compilation to NPU device specific format.
     int num_of_threads = 0;                  // [num_of_threads]: Overrides the accelerator default value of number of
                                              //  threads with this value at runtime.
     std::string load_config = "";            // Path to JSON file to load custom OV parameters.
@@ -123,10 +119,11 @@ struct OpenVINO_Provider : Provider {
     bool enable_opencl_throttling = false;   // [enable_opencl_throttling]: Enables OpenCL queue throttling for GPU
                                              // device (Reduces CPU Utilization when using GPU)
 
-    void* context = nullptr;
+    bool enable_qdq_optimizer = false;  // Enables QDQ pruning for efficient inference latency with NPU
 
-    bool enable_qdq_optimizer = false;
+    void* context = nullptr;
 
+    std::string bool_flag = "";
     if (provider_options_map.find("device_type") != provider_options_map.end()) {
       device_type = provider_options_map.at("device_type").c_str();
 
@@ -243,16 +240,6 @@ struct OpenVINO_Provider : Provider {
                               << "Executing with num_streams=1";
       }
     }
-    std::string bool_flag = "";
-    if (provider_options_map.find("enable_npu_fast_compile") != provider_options_map.end()) {
-      bool_flag = provider_options_map.at("enable_npu_fast_compile");
-      if (bool_flag == "true" || bool_flag == "True")
-        enable_npu_fast_compile = true;
-      else if (bool_flag == "false" || bool_flag == "False")
-        enable_npu_fast_compile = false;
-      bool_flag = "";
-    }
-
     if (provider_options_map.find("enable_opencl_throttling") != provider_options_map.end()) {
       bool_flag = provider_options_map.at("enable_opencl_throttling");
       if (bool_flag == "true" || bool_flag == "True")
@@ -292,11 +279,11 @@ struct OpenVINO_Provider : Provider {
           disable_dynamic_shapes = false;
         }
       }
+      bool_flag = "";
     }
 
     return std::make_shared<OpenVINOProviderFactory>(device_type,
                                                      precision,
-                                                     enable_npu_fast_compile,
                                                      num_of_threads,
                                                      load_config,
                                                      cache_dir,

onnxruntime/core/session/provider_bridge_ort.cc

@@ -1788,12 +1788,6 @@ ProviderOptions OrtOpenVINOProviderOptionsToOrtOpenVINOProviderOptionsV2(const O
   if (legacy_ov_options->device_type != nullptr)
     ov_options_converted_map["device_type"] = legacy_ov_options->device_type;
 
-  if (legacy_ov_options->enable_npu_fast_compile) {
-    ov_options_converted_map["enable_npu_fast_compile"] = "false";
-  } else {
-    ov_options_converted_map["enable_npu_fast_compile"] = "true";
-  }
-
   if (legacy_ov_options->num_of_threads != '\0')
     ov_options_converted_map["num_of_threads"] = std::to_string(legacy_ov_options->num_of_threads);
 

onnxruntime/python/onnxruntime_pybind_state.cc

@@ -1062,12 +1062,6 @@ std::unique_ptr<IExecutionProvider> CreateExecutionProviderInstance(
         } else if (option.first == "precision") {
           OV_provider_options_map[option.first] = option.second;
           continue;
-        } else if (option.first == "enable_npu_fast_compile") {
-          if (!(option.second == "True" || option.second == "true" ||
-                option.second == "False" || option.second == "false")) {
-            ORT_THROW("Invalid value passed for enable_npu_fast_compile: ", option.second);
-          }
-          OV_provider_options_map[option.first] = option.second;
         } else if (option.first == "enable_opencl_throttling") {
           if (!(option.second == "True" || option.second == "true" ||
                 option.second == "False" || option.second == "false")) {
@@ -1112,9 +1106,6 @@ std::unique_ptr<IExecutionProvider> CreateExecutionProviderInstance(
         } else if (option.first == "context") {
           OV_provider_options_map[option.first] = option.second;
           continue;
-        } else if (option.first == "export_ep_ctx_blob") {
-          OV_provider_options_map[option.first] = option.second;
-          continue;
         } else if (option.first == "enable_qdq_optimizer") {
           OV_provider_options_map[option.first] = option.second;
           continue;

onnxruntime/test/perftest/command_args_parser.cc

@@ -76,11 +76,10 @@ namespace perftest {
       "\n"
       "\t    [OpenVINO only] [device_type]: Overrides the accelerator hardware type and precision with these values at runtime.\n"
       "\t    [OpenVINO only] [device_id]: Selects a particular hardware device for inference.\n"
-      "\t    [OpenVINO only] [enable_npu_fast_compile]: Optionally enabled to speeds up the model's compilation on NPU device targets.\n"
       "\t    [OpenVINO only] [num_of_threads]: Overrides the accelerator hardware type and precision with these values at runtime.\n"
       "\t    [OpenVINO only] [cache_dir]: Explicitly specify the path to dump and load the blobs(Model caching) or cl_cache (Kernel Caching) files feature. If blob files are already present, it will be directly loaded.\n"
       "\t    [OpenVINO only] [enable_opencl_throttling]: Enables OpenCL queue throttling for GPU device(Reduces the CPU Utilization while using GPU) \n"
-      "\t    [Example] [For OpenVINO EP] -e openvino -i \"device_type|CPU enable_npu_fast_compile|true num_of_threads|5 enable_opencl_throttling|true cache_dir|\"<path>\"\"\n"
+      "\t    [Example] [For OpenVINO EP] -e openvino -i \"device_type|CPU num_of_threads|5 enable_opencl_throttling|true cache_dir|\"<path>\"\"\n"
       "\n"
       "\t    [QNN only] [backend_path]: QNN backend path. e.g '/folderpath/libQnnHtp.so', '/folderpath/libQnnCpu.so'.\n"
       "\t    [QNN only] [profiling_level]: QNN profiling level, options: 'basic', 'detailed', default 'off'.\n"

onnxruntime/test/perftest/ort_test_session.cc

@@ -795,13 +795,6 @@ select from 'TF8', 'TF16', 'UINT8', 'FLOAT', 'ITENSOR'. \n)");
             ORT_THROW("[ERROR] [OpenVINO] Unsupported inference precision is selected. CPU only supports FP32 . \n");
           }
         }
-      } else if (key == "enable_npu_fast_compile") {
-        if (value == "true" || value == "True" ||
-            value == "false" || value == "False") {
-          ov_options[key] = value;
-        } else {
-          ORT_THROW("[ERROR] [OpenVINO] The value for the key 'enable_npu_fast_compile' should be a boolean i.e. true or false. Default value is false.\n");
-        }
       } else if (key == "enable_opencl_throttling") {
         if (value == "true" || value == "True" ||
             value == "false" || value == "False") {
@@ -845,19 +838,13 @@ select from 'TF8', 'TF16', 'UINT8', 'FLOAT', 'ITENSOR'. \n)");
         } else {
           ov_options[key] = value;
         }
-      } else if (key == "export_ep_ctx_blob") {
-        if (value == "true" || value == "True" ||
-            value == "false" || value == "False") {
-          ov_options[key] = value;
-        } else {
-          ORT_THROW(
-              "[ERROR] [OpenVINO] The value for the key 'export_ep_ctx_blob' "
-              "should be a boolean i.e. true or false. Default value is false.\n");
-        }
       } else if (key == "device_memory_name") {
         device_memory_name_ = std::move(value);
       } else {
-        ORT_THROW("[ERROR] [OpenVINO] wrong key type entered. Choose from the following runtime key options that are available for OpenVINO. ['device_type', 'device_id', 'enable_npu_fast_compile', 'num_of_threads', 'load_config', 'cache_dir', 'num_streams', 'enable_opencl_throttling', 'disable_dynamic_shapes'] \n");
+        ORT_THROW(
+            "[ERROR] [OpenVINO] wrong key type entered. Choose from the following runtime key options that are available for OpenVINO."
+            " ['device_type', 'device_id', 'num_of_threads', 'load_config', 'cache_dir', 'num_streams', "
+            "'enable_opencl_throttling', 'disable_dynamic_shapes', 'enable_qdq_optimizer', 'model_priority'] \n");
       }
     }
     session_options.AppendExecutionProvider_OpenVINO_V2(ov_options);

onnxruntime/test/providers/cpu/model_tests.cc

@@ -570,7 +570,7 @@ ::std::vector<::std::basic_string<ORTCHAR_T>> GetParameterStrings() {
       ORT_TSTR("yolov3"),
       ORT_TSTR("LSTM_Seq_lens_unpacked"),
       ORT_TSTR("tinyyolov3"),
-      //ORT_TSTR("faster_rcnn"),
+      // ORT_TSTR("faster_rcnn"),
       ORT_TSTR("mask_rcnn"),
       ORT_TSTR("coreml_FNS-Candy_ImageNet"),
       ORT_TSTR("tf_mobilenet_v2_1.0_224"),
@@ -581,7 +581,7 @@ ::std::vector<::std::basic_string<ORTCHAR_T>> GetParameterStrings() {
       ORT_TSTR("mlperf_ssd_resnet34_1200"),
       ORT_TSTR("candy"),
       ORT_TSTR("cntk_simple_seg"),
-      //ORT_TSTR("GPT2_LM_HEAD"),
+      // ORT_TSTR("GPT2_LM_HEAD"),
       ORT_TSTR("mlperf_ssd_mobilenet_300"),
       ORT_TSTR("fp16_coreml_FNS-Candy"),
       ORT_TSTR("fp16_test_tiny_yolov2"),