Add SetEpDynamicOptions and remove workload_type from run/session options

.vscode/settings.json

@@ -22,5 +22,8 @@
         "-build/include_subdir",
         "-runtime/references"
     ],
-    "C_Cpp.autoAddFileAssociations": false
+    "C_Cpp.autoAddFileAssociations": false,
+    "githubPullRequests.ignoredPullRequestBranches": [
+        "main"
+    ]
 }

include/onnxruntime/core/framework/execution_provider.h

@@ -214,6 +214,15 @@ class IExecutionProvider {
     return Status::OK();
   }
 
+  /**
+     Called when InferenceSession::SetEpDynamicOptions is called
+     TODO: what is the right way of passing parameters?
+  */
+  virtual common::Status SetEpDynamicOptions(gsl::span<const char*> /*keys*/,
+                                             gsl::span<const char*> /*values*/) {
+    return Status::OK();
+  }
+
   /**
      Indicate whether the graph capturing mode (e.g., cuda graph) is enabled for
      the provider.

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -4722,6 +4722,24 @@ struct OrtApi {
    * \param[in] adapter OrtLoraAdapter instance
    */
   ORT_API2_STATUS(RunOptionsAddActiveLoraAdapter, _Inout_ OrtRunOptions* options, _In_ const OrtLoraAdapter* adapter);
+
+  /// @}
+  /// \name OrtEpDynamicOptions
+  /// @{
+
+  /** \brief Set DynamicOptions for EPs
+   *
+   * \param[in] session
+   * \param[in] list of keys represented by null-terminated strings
+   * \param[in] list of values represented by null-terminated strings
+   * \param[in] number of key-value pairs
+   *
+   * \since Version xxx
+   * @TODO: update version number
+   * @TODO: should it be SetExecutionProvider... instead of SetEp...?
+   */
+  ORT_API2_STATUS(SetEpDynamicOptions, _In_ OrtSession* sess, _In_ const char** keys, _In_ const char** values,
+                  _In_ size_t kv_len);
 };
 
 /*

include/onnxruntime/core/session/onnxruntime_run_options_config_keys.h

@@ -49,8 +49,3 @@ static const char* const kOrtRunOptionsConfigQnnRpcControlLatency = "qnn.rpc_con
 // If the value is set to -1, cuda graph capture/replay is disabled in that run.
 // User are not expected to set the value to 0 as it is reserved for internal use.
 static const char* const kOrtRunOptionsConfigCudaGraphAnnotation = "gpu_graph_id";
-
-// Specify the type of workload for this run.
-// “Default”: OS determines the scheduling priority and processor performance to service this workload. [Default]
-// “Efficient”: OS treats this workload is efficiency oriented with low scheduling priority and efficient processor performance.
-static const char* const kOrtRunOptionsWorkloadType = "run.workload_type";

include/onnxruntime/core/session/onnxruntime_session_options_config_keys.h

@@ -282,8 +282,3 @@ static const char* const kOrtSessionOptionsMlasGemmFastMathArm64Bfloat16 = "mlas
 // Refer to MatMulNBits op schema for more details.
 // If not provided, default is 4.
 static const char* const kOrtSessionOptionsQDQMatMulNBitsAccuracyLevel = "session.qdq_matmulnbits_accuracy_level";
-
-// Specify the type of workload for this session.
-// “Default”: OS determines the scheduling priority and processor performance to service this workload. [Default]
-// “Efficient”: OS treats this workload is efficiency oriented with low scheduling priority and efficient processor performance.
-static const char* const kOrtSessionOptionsWorkloadType = "session.workload_type";

onnxruntime/core/session/inference_session.cc

@@ -2475,6 +2475,38 @@
 };
 }  // namespace
 
+// TODO: are we going to do all the ttracing and logging and telemetry for this?
+// TODO: is this the right type? How do we convert to string?
+Status InferenceSession::SetEpDynamicOptions(gsl::span<const char*> keys,
+                                             gsl::span<const char*> values) {
+  Status retval = Status::OK();
+
+  ORT_TRY {
+    if (!is_inited_) {
+      LOGS(*session_logger_, ERROR) << "Session was not initialized";
+      return Status(common::ONNXRUNTIME, common::FAIL, "Session not initialized.");
+    }
+
+    // info all execution providers InferenceSession:Run started
+    // TODO: only call SetEpDynamicOptions for all providers in-use
+    for (auto& xp : execution_providers_) {
+      auto status = xp->SetEpDynamicOptions(keys, values);
+      ORT_CHECK_AND_SET_RETVAL(status);
+    }
+  }
+
+  ORT_CATCH(const std::exception& e) {
+    ORT_HANDLE_EXCEPTION([&]() {
+      retval = Status(common::ONNXRUNTIME, common::FAIL, e.what());
+    });
+  }
+  ORT_CATCH(...) {
+    retval = Status(common::ONNXRUNTIME, common::RUNTIME_EXCEPTION, "Encountered unknown exception in SetEpDynamicOptions()");
+  }
+
+  return retval;
+}
+
 Status InferenceSession::Run(const RunOptions& run_options,
                              gsl::span<const std::string> feed_names, gsl::span<const OrtValue> feeds,
                              gsl::span<const std::string> output_names, std::vector<OrtValue>* p_fetches,

onnxruntime/core/session/inference_session.h

@@ -330,6 +330,9 @@ class InferenceSession {
    */
   [[nodiscard]] common::Status Initialize();
 
+  [[nodiscard]] common::Status SetEpDynamicOptions(gsl::span<const char*> keys,
+                                                   gsl::span<const char*> values);
+
   [[nodiscard]] common::Status Run(const RunOptions& run_options, gsl::span<const std::string> feed_names,
                                    gsl::span<const OrtValue> feeds, gsl::span<const std::string> output_names,
                                    std::vector<OrtValue>* p_fetches,

onnxruntime/core/session/onnxruntime_c_api.cc

@@ -843,6 +843,28 @@
 
 }  // namespace
 
+ORT_API_STATUS_IMPL(OrtApis::SetEpDynamicOptions, _In_ OrtSession* sess, _In_ const char** keys,
+                  _In_ const char** values, _In_ size_t kv_len) {
+  API_IMPL_BEGIN
+  auto session = reinterpret_cast<::onnxruntime::InferenceSession*>(sess);
+
+  // TODO: is this what we're supposed to do with strings?
+  auto keys_span = gsl::make_span(keys, kv_len);
+  auto values_span = gsl::make_span(values, kv_len);
+
+  Status status;
+
+  if (kv_len == 0) {
+    // TODO: how does one return OK or no_values_passed
+    status = Status::OK();
+  } else {
+    status = session->SetEpDynamicOptions(keys_span,
+                                          values_span);
+  }
+  return ToOrtStatus(status);
+  API_IMPL_END
+}
+
 ORT_API_STATUS_IMPL(OrtApis::Run, _Inout_ OrtSession* sess, _In_opt_ const OrtRunOptions* run_options,
                     _In_reads_(input_len) const char* const* input_names,
                     _In_reads_(input_len) const OrtValue* const* input, size_t input_len,
@@ -2785,6 +2807,8 @@
     &OrtApis::CreateLoraAdapterFromArray,
     &OrtApis::ReleaseLoraAdapter,
     &OrtApis::RunOptionsAddActiveLoraAdapter,
+
+    &OrtApis::SetEpDynamicOptions,
 };
 
 // OrtApiBase can never change as there is no way to know what version of OrtApiBase is returned by OrtGetApiBase.

onnxruntime/core/session/ort_apis.h

@@ -531,4 +531,6 @@
 ORT_API(void, ReleaseLoraAdapter, _Frees_ptr_opt_ OrtLoraAdapter*);
 ORT_API_STATUS_IMPL(RunOptionsAddActiveLoraAdapter, _Inout_ OrtRunOptions* options, _In_ const OrtLoraAdapter* adapter);
 
+ORT_API_STATUS_IMPL(SetEpDynamicOptions, _In_ OrtSession* sess, _In_ const char** keys, _In_ const char** values,
+                  _In_ size_t kv_len);
 }  // namespace OrtApis