add used_from to inference_session

onnxruntime/core/framework/session_options.h

@@ -226,6 +226,9 @@ struct SessionOptions {
   bool has_explicit_ep_context_gen_options = false;
   epctx::ModelGenOptions ep_context_gen_options = {};
   epctx::ModelGenOptions GetEpContextGenerationOptions() const;
+
+  // Optional field to indicate where the session was used from. One of compilation, empty.
+  std::string used_from;
 };
 
 inline std::ostream& operator<<(std::ostream& os, const SessionOptions& session_options) {

onnxruntime/core/platform/telemetry.cc

@@ -62,7 +62,7 @@ void Telemetry::LogSessionCreation(uint32_t session_id, int64_t ir_version, cons
                                    const std::string& model_weight_hash,
                                    const std::unordered_map<std::string, std::string>& model_metadata,
                                    const std::string& loadedFrom, const std::vector<std::string>& execution_provider_ids,
-                                   bool use_fp16, bool captureState) const {
+                                   bool use_fp16, bool captureState, const std::string& used_from) const {
   ORT_UNUSED_PARAMETER(session_id);
   ORT_UNUSED_PARAMETER(ir_version);
   ORT_UNUSED_PARAMETER(model_producer_name);
@@ -79,6 +79,7 @@ void Telemetry::LogSessionCreation(uint32_t session_id, int64_t ir_version, cons
   ORT_UNUSED_PARAMETER(execution_provider_ids);
   ORT_UNUSED_PARAMETER(use_fp16);
   ORT_UNUSED_PARAMETER(captureState);
+  ORT_UNUSED_PARAMETER(used_from);
 }
 
 void Telemetry::LogRuntimeError(uint32_t session_id, const common::Status& status, const char* file,

onnxruntime/core/platform/telemetry.h

@@ -64,7 +64,7 @@ class Telemetry {
                                   const std::string& model_weight_hash,
                                   const std::unordered_map<std::string, std::string>& model_metadata,
                                   const std::string& loadedFrom, const std::vector<std::string>& execution_provider_ids,
-                                  bool use_fp16, bool captureState) const;
+                                  bool use_fp16, bool captureState, const std::string& used_from) const;
 
   virtual void LogRuntimeError(uint32_t session_id, const common::Status& status, const char* file,
                                const char* function, uint32_t line) const;

onnxruntime/core/platform/windows/telemetry.cc

@@ -235,7 +235,7 @@ void WindowsTelemetry::LogSessionCreation(uint32_t session_id, int64_t ir_versio
                                           const std::string& model_weight_hash,
                                           const std::unordered_map<std::string, std::string>& model_metadata,
                                           const std::string& loaded_from, const std::vector<std::string>& execution_provider_ids,
-                                          bool use_fp16, bool captureState) const {
+                                          bool use_fp16, bool captureState, const std::string& used_from) const {
   if (global_register_count_ == 0 || enabled_ == false)
     return;
 
@@ -304,7 +304,8 @@ void WindowsTelemetry::LogSessionCreation(uint32_t session_id, int64_t ir_versio
                       TraceLoggingString(model_weight_hash.c_str(), "modelWeightHash"),
                       TraceLoggingString(model_metadata_string.c_str(), "modelMetaData"),
                       TraceLoggingString(loaded_from.c_str(), "loadedFrom"),
-                      TraceLoggingString(execution_provider_string.c_str(), "executionProviderIds"));
+                      TraceLoggingString(execution_provider_string.c_str(), "executionProviderIds"),
+                      TraceLoggingString(used_from.c_str(), "usedFrom"));
   } else {
     TraceLoggingWrite(telemetry_provider_handle,
                       "SessionCreation_CaptureState",
@@ -330,7 +331,8 @@ void WindowsTelemetry::LogSessionCreation(uint32_t session_id, int64_t ir_versio
                       TraceLoggingString(model_weight_hash.c_str(), "modelWeightHash"),
                       TraceLoggingString(model_metadata_string.c_str(), "modelMetaData"),
                       TraceLoggingString(loaded_from.c_str(), "loadedFrom"),
-                      TraceLoggingString(execution_provider_string.c_str(), "executionProviderIds"));
+                      TraceLoggingString(execution_provider_string.c_str(), "executionProviderIds"),
+                      TraceLoggingString(used_from.c_str(), "usedFrom"));
   }
 }
 

onnxruntime/core/platform/windows/telemetry.h

@@ -57,7 +57,7 @@ class WindowsTelemetry : public Telemetry {
                           const std::string& model_weight_hash,
                           const std::unordered_map<std::string, std::string>& model_metadata,
                           const std::string& loadedFrom, const std::vector<std::string>& execution_provider_ids,
-                          bool use_fp16, bool captureState) const override;
+                          bool use_fp16, bool captureState, const std::string& used_from) const override;
 
   void LogRuntimeError(uint32_t session_id, const common::Status& status, const char* file,
                        const char* function, uint32_t line) const override;

onnxruntime/core/session/inference_session.cc

@@ -518,6 +518,7 @@
   }
 
   telemetry_ = {};
+  telemetry_.used_from = session_options_.used_from;
 
 #ifdef _WIN32
   std::lock_guard<std::mutex> lock(active_sessions_mutex_);
@@ -2504,7 +2505,7 @@
     env.GetTelemetryProvider().LogSessionCreation(
         session_id_, model_->IrVersion(), model_->ProducerName(), model_->ProducerVersion(), model_->Domain(),
         graph.DomainToVersionMap(), model_file_name, graph.Name(), model_weight_type, model_graph_hash, model_weight_hash,
-        model_->MetaData(), telemetry_.event_name_, execution_providers_.GetIds(), model_has_fp16_inputs, false);
+        model_->MetaData(), telemetry_.event_name_, execution_providers_.GetIds(), model_has_fp16_inputs, false, telemetry_.used_from_);
 
     LOGS(*session_logger_, INFO) << "Session successfully initialized.";
   }
@@ -3890,7 +3891,7 @@
       std::string model_weight_type = session->GetWeightDataType();
       std::string model_graph_hash = session->GetGraphHash();
       std::string model_weight_hash = session->GetWeightHash();
       env.GetTelemetryProvider().LogSessionCreation(
           session->session_id_, model->IrVersion(), model->ProducerName(), model->ProducerVersion(), model->Domain(),
           graph.DomainToVersionMap(), model_file_name, graph.Name(), model_weight_type, model_graph_hash, model_weight_hash,
           model->MetaData(), session->telemetry_.event_name_, session->execution_providers_.GetIds(), model_has_fp16_inputs, true);

onnxruntime/core/session/inference_session.h

@@ -944,6 +944,7 @@ class InferenceSession {
     uint32_t total_runs_since_last_ = 0;                              // the total number of Run() calls since the last report
     long long total_run_duration_since_last_ = 0;                     // the total duration (us) of Run() calls since the last report
     std::string event_name_;                                          // where the model is loaded from: ["model_loading_uri", "model_loading_proto", "model_loading_istream"]
+    std::string used_from_;
     std::unordered_map<int64_t, long long> duration_per_batch_size_;  // the duration (us) of Run() calls per batch size since the last report
 
     TimePoint time_sent_last_;  // the TimePoint of the last report

onnxruntime/core/session/model_compilation_options.cc

@@ -18,6 +18,7 @@
 namespace onnxruntime {
 ModelCompilationOptions::ModelCompilationOptions(const onnxruntime::Environment& env, const OrtSessionOptions& session_options)
     : env_(env), session_options_(session_options) {
+  session_options_.value.used_from = "compilation";
   session_options_.value.has_explicit_ep_context_gen_options = true;
   session_options_.value.ep_context_gen_options = session_options.value.GetEpContextGenerationOptions();
   session_options_.value.ep_context_gen_options.enable = true;