Initial draft

Author: Aditya Rastogi

csharp/src/Microsoft.ML.OnnxRuntime/NativeCompileApiMethods.shared.cs

@@ -25,6 +25,7 @@ public struct OrtCompileApi
         public IntPtr ModelCompilationOptions_SetGraphOptimizationLevel;
         public IntPtr ModelCompilationOptions_SetOutputModelWriteFunc;
         public IntPtr ModelCompilationOptions_SetOutputModelGetInitializerLocationFunc;
+        public IntPtr ModelCompilationOptions_SetInputModel;
     }
 
     internal class NativeMethods
@@ -136,6 +137,12 @@ public DOrtModelCompilationOptions_SetOutputModelWriteFunc
         public DOrtModelCompilationOptions_SetOutputModelGetInitializerLocationFunc
                         OrtModelCompilationOptions_SetOutputModelGetInitializerLocationFunc;
 
+        [UnmanagedFunctionPointer(CallingConvention.Winapi)]
+        public delegate IntPtr /* OrtStatus* */ DOrtModelCompilationOptions_SetInputModel(
+            IntPtr /* OrtModelCompilationOptions* */ options,
+            IntPtr /* const OrtModel* */ inputModel);
+        public DOrtModelCompilationOptions_SetInputModel OrtModelCompilationOptions_SetInputModel;
+
         internal NativeMethods(OnnxRuntime.NativeMethods.DOrtGetCompileApi getCompileApi)
         {
 
@@ -217,6 +224,11 @@ internal NativeMethods(OnnxRuntime.NativeMethods.DOrtGetCompileApi getCompileApi
                     _compileApi.ModelCompilationOptions_SetOutputModelGetInitializerLocationFunc,
                     typeof(DOrtModelCompilationOptions_SetOutputModelGetInitializerLocationFunc));
 
+            OrtModelCompilationOptions_SetInputModel =
+                (DOrtModelCompilationOptions_SetInputModel)Marshal.GetDelegateForFunctionPointer(
+                    _compileApi.ModelCompilationOptions_SetInputModel,
+                    typeof(DOrtModelCompilationOptions_SetInputModel));
+
         }
     }
 }

include/onnxruntime/core/session/onnxruntime_c_api.h

@@ -8037,6 +8037,29 @@ struct OrtCompileApi {
   ORT_API2_STATUS(ModelCompilationOptions_SetOutputModelGetInitializerLocationFunc,
                   _In_ OrtModelCompilationOptions* model_compile_options,
                   _In_ OrtGetInitializerLocationFunc get_initializer_location_func, _In_ void* state);
+
+  /** \brief Sets the OrtModel to compile.
+   *
+   * Sets an OrtModel created via the Model Editor API as the input for compilation.
+   *
+   * The input model's source (file path, memory buffer, or OrtModel) must be set with
+   * one of: ModelCompilationOptions_SetInputModelPath, ModelCompilationOptions_SetInputModelFromBuffer,
+   * or ModelCompilationOptions_SetInputModel.
+   *
+   * The OrtModel must have a complete graph with inputs, outputs, and nodes defined.
+   * The caller retains ownership of the OrtModel and must not release it until after
+   * CompileModel returns.
+   *
+   * \param[in] model_compile_options The OrtModelCompilationOptions instance.
+   * \param[in] model The OrtModel to compile. The model is borrowed (not copied or owned).
+   *
+   * \snippet{doc} snippets.dox OrtStatus Return Value
+   *
+   * \since Version 1.24.
+   */
+  ORT_API2_STATUS(ModelCompilationOptions_SetInputModel,
+                  _In_ OrtModelCompilationOptions* model_compile_options,
+                  _In_ const OrtModel* model);
 };
 
 /**

include/onnxruntime/core/session/onnxruntime_cxx_api.h

@@ -1612,6 +1612,8 @@ struct ModelCompilationOptions : detail::Base<OrtModelCompilationOptions> {
   ModelCompilationOptions& SetFlags(uint32_t flags);                                    ///< Wraps OrtApi::ModelCompilationOptions_SetFlags
 
   ModelCompilationOptions& SetGraphOptimizationLevel(GraphOptimizationLevel graph_optimization_level);  ///< Wraps OrtApi::ModelCompilationOptions_SetGraphOptimizationLevel
+
+  ModelCompilationOptions& SetInputModel(const OrtModel* model);  ///< Wraps OrtCompileApi::ModelCompilationOptions_SetInputModel
 };
 
 /** \brief Compiles an input model to generate a model with EPContext nodes that execute EP-specific kernels. Wraps OrtApi::CompileModels.

include/onnxruntime/core/session/onnxruntime_cxx_inline.h

@@ -1180,6 +1180,11 @@ inline ModelCompilationOptions& ModelCompilationOptions::SetGraphOptimizationLev
   return *this;
 }
 
+inline ModelCompilationOptions& ModelCompilationOptions::SetInputModel(const OrtModel* model) {
+  Ort::ThrowOnError(GetCompileApi().ModelCompilationOptions_SetInputModel(this->p_, model));
+  return *this;
+}
+
 namespace detail {
 
 template <typename T>

onnxruntime/core/session/compile_api.cc

@@ -306,6 +306,27 @@ ORT_API_STATUS_IMPL(OrtCompileAPI::ModelCompilationOptions_SetGraphOptimizationL
   API_IMPL_END
 }
 
+ORT_API_STATUS_IMPL(OrtCompileAPI::ModelCompilationOptions_SetInputModel,
+                    _In_ OrtModelCompilationOptions* ort_model_compile_options,
+                    _In_ const OrtModel* model) {
+  API_IMPL_BEGIN
+#if !defined(ORT_MINIMAL_BUILD)
+  auto model_compile_options = reinterpret_cast<onnxruntime::ModelCompilationOptions*>(ort_model_compile_options);
+
+  if (model == nullptr) {
+    return OrtApis::CreateStatus(ORT_INVALID_ARGUMENT, "Invalid input model: OrtModel pointer is null");
+  }
+
+  model_compile_options->SetInputModel(model);
+  return nullptr;
+#else
+  ORT_UNUSED_PARAMETER(ort_model_compile_options);
+  ORT_UNUSED_PARAMETER(model);
+  return OrtApis::CreateStatus(ORT_NOT_IMPLEMENTED, "Compile API is not supported in this build");
+#endif  // !defined(ORT_MINIMAL_BUILD)
+  API_IMPL_END
+}
+
 ORT_API_STATUS_IMPL(OrtCompileAPI::CompileModel, _In_ const OrtEnv* env,
                     _In_ const OrtModelCompilationOptions* ort_model_compile_options) {
   API_IMPL_BEGIN
@@ -343,13 +364,18 @@ static constexpr OrtCompileApi ort_compile_api = {
     &OrtCompileAPI::ModelCompilationOptions_SetOutputModelWriteFunc,
     &OrtCompileAPI::ModelCompilationOptions_SetOutputModelGetInitializerLocationFunc,
     // End of Version 23 - DO NOT MODIFY ABOVE
+
+    &OrtCompileAPI::ModelCompilationOptions_SetInputModel,
+    // End of Version 24 - DO NOT MODIFY ABOVE
 };
 
 // checks that we don't violate the rule that the functions must remain in the slots they were originally assigned
 static_assert(offsetof(OrtCompileApi, CompileModel) / sizeof(void*) == 8,
               "Size of version 22 Api cannot change");  // initial version in ORT 1.22
 static_assert(offsetof(OrtCompileApi, ModelCompilationOptions_SetOutputModelGetInitializerLocationFunc) / sizeof(void*) == 13,
               "Size of version 23 of Api cannot change");
+static_assert(offsetof(OrtCompileApi, ModelCompilationOptions_SetInputModel) / sizeof(void*) == 14,
+              "Size of version 24 of Api cannot change");
 
 ORT_API(const OrtCompileApi*, OrtCompileAPI::GetCompileApi) {
   return &ort_compile_api;

onnxruntime/core/session/compile_api.h

@@ -41,5 +41,8 @@ ORT_API_STATUS_IMPL(ModelCompilationOptions_SetOutputModelWriteFunc,
 ORT_API_STATUS_IMPL(ModelCompilationOptions_SetOutputModelGetInitializerLocationFunc,
                     _In_ OrtModelCompilationOptions* model_compile_options,
                     _In_ OrtGetInitializerLocationFunc get_initializer_location_func, _In_ void* state);
+ORT_API_STATUS_IMPL(ModelCompilationOptions_SetInputModel,
+                    _In_ OrtModelCompilationOptions* model_compile_options,
+                    _In_ const OrtModel* model);
 
 }  // namespace OrtCompileAPI

onnxruntime/core/session/model_compilation_options.cc

@@ -45,6 +45,11 @@ void ModelCompilationOptions::SetInputModelFromBuffer(const void* input_model_da
   input_model_data_size_ = input_model_data_size;
 }
 
+void ModelCompilationOptions::SetInputModel(const OrtModel* model) {
+  ResetInputModelSettings();
+  input_model_ = model;
+}
+
 Status ModelCompilationOptions::SetOutputModelPath(const std::filesystem::path& output_model_path) {
   ConfigOptions& config_options = session_options_.value.config_options;
   epctx::ModelGenOptions& ep_context_gen_options = session_options_.value.ep_context_gen_options;
@@ -186,10 +191,19 @@ size_t ModelCompilationOptions::GetInputModelDataSize() const {
   return input_model_data_size_;
 }
 
+bool ModelCompilationOptions::InputModelComesFromOrtModel() const {
+  return input_model_ != nullptr;
+}
+
+const OrtModel* ModelCompilationOptions::GetInputModel() const {
+  return input_model_;
+}
+
 void ModelCompilationOptions::ResetInputModelSettings() {
   input_model_path_.clear();
   input_model_data_ = nullptr;
   input_model_data_size_ = 0;
+  input_model_ = nullptr;
 }
 
 Status ModelCompilationOptions::SetGraphOptimizationLevel(GraphOptimizationLevel graph_optimization_level) {
@@ -229,16 +243,21 @@ Status ModelCompilationOptions::Check() const {
   // Check input model settings.
   const bool input_from_file = !input_model_path_.empty();
   const bool input_from_memory = input_model_data_ != nullptr;
+  const bool input_from_model = input_model_ != nullptr;
+
+  int input_source_count = (input_from_file ? 1 : 0) +
+                           (input_from_memory ? 1 : 0) +
+                           (input_from_model ? 1 : 0);
 
-  if (!input_from_file && !input_from_memory) {
+  if (input_source_count == 0) {
     return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
-                           "Input model to compile must be loaded from either a file or a memory buffer");
+                           "Input model to compile must be specified via file path, memory buffer, or OrtModel");
   }
 
-  if (input_from_file && input_from_memory) {
+  if (input_source_count > 1) {
     return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
-                           "Input model to compile must be loaded from either a file or a memory buffer, ",
-                           "but not both.");
+                           "Input model to compile must be specified via exactly one of: ",
+                           "file path, memory buffer, or OrtModel");
   }
 
   if (input_from_file && !std::filesystem::exists(input_model_path_)) {
@@ -249,12 +268,59 @@ Status ModelCompilationOptions::Check() const {
     return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT, "Buffer for input model data has size 0");
   }
 
+  // Validate OrtModel input
+  if (input_from_model) {
+    if (input_model_->graph == nullptr) {
+      return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
+                             "OrtModel has no graph. Call AddGraphToModel before compilation.");
+    }
+
+    if (input_model_->graph->GetNumInputs() == 0 || input_model_->graph->GetNumOutputs() == 0) {
+      return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
+                             "OrtModel graph must have at least one input and one output defined.");
+    }
+
+    if (input_model_->domain_to_version.empty()) {
+      return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
+                             "OrtModel must specify at least one opset domain/version.");
+    }
+
+    // Note: Additional validation (node connections, schema) happens during
+    // Model::LoadFromModelEditorApiModel -> Graph::Resolve()
+  }
+
   // Check output model settings.
   const epctx::ModelGenOptions& ep_context_gen_options = session_options_.value.ep_context_gen_options;
   bool has_no_output_model_location = std::holds_alternative<std::monostate>(
       ep_context_gen_options.output_model_location);
 
-  if (has_no_output_model_location && input_from_file) {
+  // Determine if we can derive an output path from the input
+  bool can_derive_output_path = input_from_file;
+  bool model_has_path = false;
+
+  // For OrtModel input, check if model_path is set in the graph using the virtual GetModelPath() method
+  // (avoids dynamic_cast which requires RTTI)
+  if (input_from_model && input_model_->graph) {
+    const ORTCHAR_T* model_path_cstr = input_model_->graph->GetModelPath();
+    if (model_path_cstr && model_path_cstr[0] != ORT_TSTR('\0')) {
+      can_derive_output_path = true;
+      model_has_path = true;
+    }
+  }
+
+  // Fast-fail: If OrtModel has no model_path and user hasn't specified output location or embed mode,
+  // EPs that need to write context binaries will fail later. Fail early with a clear error.
+  if (input_from_model && !model_has_path && has_no_output_model_location &&
+      !ep_context_gen_options.embed_ep_context_in_model) {
+    return ORT_MAKE_STATUS(ONNXRUNTIME, INVALID_ARGUMENT,
+                           "OrtModel has no model_path set and no output location was specified. "
+                           "Please either: (1) set the model_path on the OrtGraph before adding to OrtModel, "
+                           "(2) call SetOutputModelPath/SetOutputModelBuffer to specify an output location, "
+                           "(3) call SetEpContextEmbedMode(true) to embed EP context in the model, or "
+                           "(4) call SetEpContextBinaryInformation to specify the binary output directory.");
+  }
+
+  if (has_no_output_model_location && can_derive_output_path) {
     // User did not specify an output file, an output buffer, or an output write function. We default to generating an
     // output file with a name based on the input file name, so do not return an error.
     return Status::OK();

onnxruntime/core/session/model_compilation_options.h

@@ -10,6 +10,7 @@
 #include "core/common/status.h"
 #include "core/common/path_string.h"
 #include "core/framework/allocator.h"
+#include "core/graph/model_editor_api_types.h"
 #include "core/session/abi_session_options_impl.h"
 #include "core/session/onnxruntime_c_api.h"
 #include "core/session/onnxruntime_session_options_config_keys.h"
@@ -45,6 +46,14 @@ class ModelCompilationOptions {
   /// <param name="input_model_data_size">The size in bytes of the input model's buffer</param>
   void SetInputModelFromBuffer(const void* input_model_data, size_t input_model_data_size);
 
+  /// <summary>
+  /// Sets the OrtModel to compile.
+  /// The OrtModel is borrowed (not copied) - caller must keep it alive until CompileModel returns.
+  /// Overrides any previous call to SetInputModelPath(), SetInputModelFromBuffer(), or SetInputModel().
+  /// </summary>
+  /// <param name="model">The OrtModel to compile</param>
+  void SetInputModel(const OrtModel* model);
+
   /// <summary>
   /// Sets the file path to store the output/compiled ONNX model.
   /// Overrides any previous call to SetOutputModelPath() or SetOutputModelBuffer().
@@ -132,6 +141,18 @@ class ModelCompilationOptions {
   /// <returns>true if input model comes from a file</returns>
   bool InputModelComesFromFile() const;
 
+  /// <summary>
+  /// Returns true if the input model comes from an OrtModel pointer.
+  /// </summary>
+  /// <returns>true if input model comes from an OrtModel</returns>
+  bool InputModelComesFromOrtModel() const;
+
+  /// <summary>
+  /// Returns the OrtModel to compile, or nullptr if not set.
+  /// </summary>
+  /// <returns>pointer to the OrtModel or nullptr</returns>
+  const OrtModel* GetInputModel() const;
+
   /// <summary>
   /// Returns the buffer that contains the bytes for the input ONNX model.
   /// Returns nullptr if the input model is not stored in a buffer.
@@ -205,6 +226,7 @@ class ModelCompilationOptions {
   std::filesystem::path input_model_path_;
   const void* input_model_data_ = nullptr;
   size_t input_model_data_size_ = 0;
+  const OrtModel* input_model_ = nullptr;  // Borrowed pointer
 };
 }  // namespace onnxruntime
 #endif  // !defined(ORT_MINIMAL_BUILD)