Add support for selection policy delegate (based on PR #24653) (#24638)

### Description
Adds support for selection policy delegate directly to the release
branch. This is necessary to avoid having to update C# bindings (which
are in main but not in the release branch)

Based on #24635

### Motivation and Context
<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->

Author: adrianlizarraga

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -438,6 +438,7 @@ typedef enum OrtExecutionProviderDevicePolicy {
  * \param max_ep_devices The maximum number of devices that can be selected in the pre-allocated array.
                          Currently the maximum is 8.
  * \param num_ep_devices The number of selected devices.
+ * \param state Opaque pointer. Required to use the delegate from other languages like C# and python.
  *
  * \return OrtStatus* Selection status. Return nullptr on success.
  *                    Use CreateStatus to provide error info. Use ORT_FAIL as the error code.
@@ -449,7 +450,8 @@ typedef OrtStatus* (*EpSelectionDelegate)(_In_ const OrtEpDevice** ep_devices,
                                           _In_opt_ const OrtKeyValuePairs* runtime_metadata,
                                           _Inout_ const OrtEpDevice** selected,
                                           _In_ size_t max_selected,
-                                          _Out_ size_t* num_selected);
+                                          _Out_ size_t* num_selected,
+                                          _In_ void* state);
 
 /** \brief Algorithm to use for cuDNN Convolution Op
  */
@@ -5127,18 +5129,30 @@ struct OrtApi {
 
   /** \brief Set the execution provider selection policy for the session.
    *
-   * Allows users to specify a device selection policy for automatic execution provider (EP) selection,
-   * or provide a delegate callback for custom selection logic.
+   * Allows users to specify a device selection policy for automatic execution provider (EP) selection.
+   * If custom selection is required please use SessionOptionsSetEpSelectionPolicyDelegate instead.
    *
    * \param[in] session_options The OrtSessionOptions instance.
    * \param[in] policy The device selection policy to use (see OrtExecutionProviderDevicePolicy).
-   * \param[in] delegate Optional delegate callback for custom selection. Pass nullptr to use the built-in policy.
    *
    * \since Version 1.22
    */
   ORT_API2_STATUS(SessionOptionsSetEpSelectionPolicy, _In_ OrtSessionOptions* session_options,
-                  _In_ OrtExecutionProviderDevicePolicy policy,
-                  _In_opt_ EpSelectionDelegate* delegate);
+                  _In_ OrtExecutionProviderDevicePolicy policy);
+
+  /** \brief Set the execution provider selection policy delegate for the session.
+   *
+   * Allows users to provide a custom device selection policy for automatic execution provider (EP) selection.
+   *
+   * \param[in] session_options The OrtSessionOptions instance.
+   * \param[in] delegate Delegate callback for custom selection.
+   * \param[in] delegate_state Optional state that will be passed to the delegate callback. nullptr if not required.
+   *
+   * \since Version 1.22
+   */
+  ORT_API2_STATUS(SessionOptionsSetEpSelectionPolicyDelegate, _In_ OrtSessionOptions* session_options,
+                  _In_ EpSelectionDelegate delegate,
+                  _In_opt_ void* delegate_state);
 
   /** \brief Get the hardware device type.
    *

include/onnxruntime/core/session/onnxruntime_cxx_api.h

@@ -1103,8 +1103,10 @@ struct SessionOptionsImpl : ConstSessionOptionsImpl<T> {
                                                  const std::unordered_map<std::string, std::string>& ep_options);
 
   /// Wraps OrtApi::SessionOptionsSetEpSelectionPolicy
-  SessionOptionsImpl& SetEpSelectionPolicy(OrtExecutionProviderDevicePolicy policy,
-                                           EpSelectionDelegate* delegate = nullptr);
+  SessionOptionsImpl& SetEpSelectionPolicy(OrtExecutionProviderDevicePolicy policy);
+
+  /// Wraps OrtApi::SessionOptionsSetEpSelectionPolicyDelegate
+  SessionOptionsImpl& SetEpSelectionPolicy(EpSelectionDelegate delegate, void* state = nullptr);
 
   SessionOptionsImpl& SetCustomCreateThreadFn(OrtCustomCreateThreadFn ort_custom_create_thread_fn);  ///< Wraps OrtApi::SessionOptionsSetCustomCreateThreadFn
   SessionOptionsImpl& SetCustomThreadCreationOptions(void* ort_custom_thread_creation_options);      ///< Wraps OrtApi::SessionOptionsSetCustomThreadCreationOptions

include/onnxruntime/core/session/onnxruntime_cxx_inline.h

@@ -1150,9 +1150,14 @@ inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::AppendExecutionProvider_V2(
 }
 
 template <typename T>
-inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::SetEpSelectionPolicy(OrtExecutionProviderDevicePolicy policy,
-                                                                          EpSelectionDelegate* delegate) {
-  ThrowOnError(GetApi().SessionOptionsSetEpSelectionPolicy(this->p_, policy, delegate));
+inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::SetEpSelectionPolicy(OrtExecutionProviderDevicePolicy policy) {
+  ThrowOnError(GetApi().SessionOptionsSetEpSelectionPolicy(this->p_, policy));
+  return *this;
+}
+
+template <typename T>
+inline SessionOptionsImpl<T>& SessionOptionsImpl<T>::SetEpSelectionPolicy(EpSelectionDelegate delegate, void* state) {
+  ThrowOnError(GetApi().SessionOptionsSetEpSelectionPolicyDelegate(this->p_, delegate, state));
   return *this;
 }
 

onnxruntime/core/framework/session_options.h

@@ -96,7 +96,8 @@ struct EpSelectionPolicy {
   // and no selection policy was explicitly provided.
   bool enable{false};
   OrtExecutionProviderDevicePolicy policy = OrtExecutionProviderDevicePolicy_DEFAULT;
-  EpSelectionDelegate* delegate{};
+  EpSelectionDelegate delegate{};
+  void* state{nullptr};  // state for the delegate
 };
 
 /**

onnxruntime/core/session/abi_session_options.cc

@@ -367,12 +367,24 @@ ORT_API_STATUS_IMPL(OrtApis::SetDeterministicCompute, _Inout_ OrtSessionOptions*
 }
 
 ORT_API_STATUS_IMPL(OrtApis::SessionOptionsSetEpSelectionPolicy, _In_ OrtSessionOptions* options,
-                    _In_ OrtExecutionProviderDevicePolicy policy,
-                    _In_opt_ EpSelectionDelegate* delegate) {
+                    _In_ OrtExecutionProviderDevicePolicy policy) {
   API_IMPL_BEGIN
   options->value.ep_selection_policy.enable = true;
   options->value.ep_selection_policy.policy = policy;
+  options->value.ep_selection_policy.delegate = nullptr;
+  options->value.ep_selection_policy.state = nullptr;
+  return nullptr;
+  API_IMPL_END
+}
+
+ORT_API_STATUS_IMPL(OrtApis::SessionOptionsSetEpSelectionPolicyDelegate, _In_ OrtSessionOptions* options,
+                    _In_opt_ EpSelectionDelegate delegate,
+                    _In_opt_ void* state) {
+  API_IMPL_BEGIN
+  options->value.ep_selection_policy.enable = true;
+  options->value.ep_selection_policy.policy = OrtExecutionProviderDevicePolicy_DEFAULT;
   options->value.ep_selection_policy.delegate = delegate;
+  options->value.ep_selection_policy.state = state;
   return nullptr;
   API_IMPL_END
 }

onnxruntime/core/session/inference_session.cc

@@ -3264,6 +3264,7 @@ common::Status InferenceSession::SaveModelMetadata(const onnxruntime::Model& mod
 
   // save model metadata
   model_metadata_.producer_name = model.ProducerName();
+  model_metadata_.producer_version = model.ProducerVersion();
   model_metadata_.description = model.DocString();
   model_metadata_.graph_description = model.GraphDocString();
   model_metadata_.domain = model.Domain();
@@ -3428,6 +3429,10 @@ const Model& InferenceSession::GetModel() const {
   return *model_;
 }
 
+const Environment& InferenceSession::GetEnvironment() const {
+  return environment_;
+}
+
 SessionIOBinding::SessionIOBinding(InferenceSession* session) : sess_(session) {
   ORT_ENFORCE(session->NewIOBinding(&binding_).IsOK());
 }

onnxruntime/core/session/inference_session.h

@@ -78,6 +78,7 @@ struct ModelMetadata {
   ModelMetadata& operator=(const ModelMetadata&) = delete;
 
   std::string producer_name;
+  std::string producer_version;
   std::string graph_name;
   std::string domain;
   std::string description;
@@ -601,6 +602,7 @@ class InferenceSession {
 #endif
 
   const Model& GetModel() const;
+  const Environment& GetEnvironment() const;
 
  protected:
 #if !defined(ORT_MINIMAL_BUILD)

onnxruntime/core/session/onnxruntime_c_api.cc

@@ -3013,6 +3013,7 @@ static constexpr OrtApi ort_api_1_to_22 = {
     &OrtApis::GetEpDevices,
     &OrtApis::SessionOptionsAppendExecutionProvider_V2,
     &OrtApis::SessionOptionsSetEpSelectionPolicy,
+    &OrtApis::SessionOptionsSetEpSelectionPolicyDelegate,
 
     &OrtApis::HardwareDevice_Type,
     &OrtApis::HardwareDevice_VendorId,
@@ -3062,7 +3063,7 @@ static_assert(offsetof(OrtApi, AddExternalInitializersFromFilesInMemory) / sizeo
 // no additions in version 19, 20, and 21
 static_assert(offsetof(OrtApi, SetEpDynamicOptions) / sizeof(void*) == 284, "Size of version 20 API cannot change");
 
-static_assert(offsetof(OrtApi, GetEpApi) / sizeof(void*) == 316, "Size of version 22 API cannot change");
+static_assert(offsetof(OrtApi, GetEpApi) / sizeof(void*) == 317, "Size of version 22 API cannot change");
 
 // So that nobody forgets to finish an API version, this check will serve as a reminder:
 static_assert(std::string_view(ORT_VERSION) == "1.22.0",

onnxruntime/core/session/ort_apis.h

@@ -576,8 +576,11 @@ ORT_API_STATUS_IMPL(SessionOptionsAppendExecutionProvider_V2, _In_ OrtSessionOpt
                     size_t num_ep_options);
 
 ORT_API_STATUS_IMPL(SessionOptionsSetEpSelectionPolicy, _In_ OrtSessionOptions* sess_options,
-                    _In_ OrtExecutionProviderDevicePolicy policy,
-                    _In_opt_ EpSelectionDelegate* delegate);
+                    _In_ OrtExecutionProviderDevicePolicy policy);
+
+ORT_API_STATUS_IMPL(SessionOptionsSetEpSelectionPolicyDelegate, _In_ OrtSessionOptions* sess_options,
+                    _In_ EpSelectionDelegate delegate,
+                    _In_opt_ void* state);
 
 // OrtHardwareDevice accessors.
 ORT_API(OrtHardwareDeviceType, HardwareDevice_Type, _In_ const OrtHardwareDevice* device);

onnxruntime/core/session/provider_policy_context.cc

@@ -94,8 +94,8 @@ std::vector<const OrtEpDevice*> OrderDevices(const std::vector<const OrtEpDevice
     bool bIsDefaultCpuEp = IsDefaultCpuEp(b);
     if (!aIsDefaultCpuEp && !bIsDefaultCpuEp) {
       // neither are default CPU EP. both do/don't match vendor.
-      // TODO: implement tie-breaker for this scenario. arbitrarily prefer the first for now
-      return true;
+      // TODO: implement tie-breaker for this scenario. arbitrarily sort by ep name
+      return a->ep_name < b->ep_name;
     }
 
     // one is the default CPU EP
@@ -104,31 +104,57 @@ std::vector<const OrtEpDevice*> OrderDevices(const std::vector<const OrtEpDevice
 
   return sorted_devices;
 }
+
+OrtKeyValuePairs GetModelMetadata(const InferenceSession& session) {
+  OrtKeyValuePairs metadata;
+  auto status_and_metadata = session.GetModelMetadata();
+
+  if (!status_and_metadata.first.IsOK()) {
+    return metadata;
+  }
+
+  // use field names from onnx.proto
+  const auto& model_metadata = *status_and_metadata.second;
+  metadata.Add("producer_name", model_metadata.producer_name);
+  metadata.Add("producer_version", model_metadata.producer_version);
+  metadata.Add("domain", model_metadata.domain);
+  metadata.Add("model_version", std::to_string(model_metadata.version));
+  metadata.Add("doc_string", model_metadata.description);
+  metadata.Add("graph_name", model_metadata.graph_name);                // name from main GraphProto
+  metadata.Add("graph_description", model_metadata.graph_description);  // descriptions from main GraphProto
+  for (const auto& entry : model_metadata.custom_metadata_map) {
+    metadata.Add(entry.first, entry.second);
+  }
+
+  return metadata;
+}
 }  // namespace
 
 // Select execution providers based on the device policy and available devices and add to session
 Status ProviderPolicyContext::SelectEpsForSession(const Environment& env, const OrtSessionOptions& options,
                                                   InferenceSession& sess) {
-  ORT_ENFORCE(options.value.ep_selection_policy.delegate == nullptr,
-              "EP selection delegate support is not implemented yet.");
-
   // Get the list of devices from the environment and order them.
   // Ordered by preference within each type. NPU -> GPU -> NPU
   // TODO: Should environment.cc do the ordering?
-  const auto& execution_devices = OrderDevices(env.GetOrtEpDevices());
+  std::vector<const OrtEpDevice*> execution_devices = OrderDevices(env.GetOrtEpDevices());
 
   // The list of devices selected by policies
   std::vector<const OrtEpDevice*> devices_selected;
 
   // Run the delegate if it was passed in lieu of any other policy
   if (options.value.ep_selection_policy.delegate) {
-    auto policy_fn = options.value.ep_selection_policy.delegate;
+    auto model_metadata = GetModelMetadata(sess);
+    OrtKeyValuePairs runtime_metadata;  // TODO: where should this come from?
+
     std::vector<const OrtEpDevice*> delegate_devices(execution_devices.begin(), execution_devices.end());
     std::array<const OrtEpDevice*, 8> selected_devices{nullptr};
-
     size_t num_selected = 0;
-    auto* status = (*policy_fn)(delegate_devices.data(), delegate_devices.size(),
-                                nullptr, nullptr, selected_devices.data(), selected_devices.size(), &num_selected);
+
+    EpSelectionDelegate delegate = options.value.ep_selection_policy.delegate;
+    auto* status = delegate(delegate_devices.data(), delegate_devices.size(),
+                            &model_metadata, &runtime_metadata,
+                            selected_devices.data(), selected_devices.size(), &num_selected,
+                            options.value.ep_selection_policy.state);
 
     // return or fall-through for both these cases
     // going with explicit failure for now so it's obvious to user what is happening
@@ -142,6 +168,12 @@ Status ProviderPolicyContext::SelectEpsForSession(const Environment& env, const
     if (num_selected == 0) {
       return ORT_MAKE_STATUS(ONNXRUNTIME, FAIL, "EP selection delegate did not select anything.");
     }
+
+    // Copy the selected devices to the output vector
+    devices_selected.reserve(num_selected);
+    for (size_t i = 0; i < num_selected; ++i) {
+      devices_selected.push_back(selected_devices[i]);
+    }
   } else {
     // Create the selector for the chosen policy
     std::unique_ptr<IEpPolicySelector> selector;