Allow EpContext models with input/output models completely in buffers (microsoft#24463)

### Description
Re-enables (and fixes) generation of compiled EpContext models with
**both** input and output models stored in buffers.

### Motivation and Context
Previous PR microsoft#24176 inadvertently added a check that disabled storing
both input and output models in buffers. However, we need this
functionality. This was actually a fortunate scenario, as it led to the
discovery of a bug.

Author: adrianlizarraga

onnxruntime/core/framework/graph_partitioner.cc

@@ -818,11 +818,17 @@ static Status CreateEpContextModel(const ExecutionProviders& execution_providers
     return std::make_pair(false, static_cast<const Node*>(nullptr));
   };
 
+  bool saving_to_buffer = ep_context_gen_options.output_model_buffer_ptr != nullptr &&
+                          ep_context_gen_options.output_model_buffer_size_ptr != nullptr &&
+                          ep_context_gen_options.output_model_buffer_allocator != nullptr;
+
   std::filesystem::path context_cache_path;
-  ORT_RETURN_IF_ERROR(GetValidatedEpContextPath(ep_context_gen_options.output_model_file_path,
-                                                graph.ModelPath(),
-                                                context_cache_path,
-                                                ep_context_gen_options.overwrite_existing_output_file));
+  if (!saving_to_buffer || !graph.ModelPath().empty()) {
+    ORT_RETURN_IF_ERROR(GetValidatedEpContextPath(ep_context_gen_options.output_model_file_path,
+                                                  graph.ModelPath(),
+                                                  context_cache_path,
+                                                  ep_context_gen_options.overwrite_existing_output_file));
+  }
 
   Model ep_context_model(graph.Name(), false, graph.GetModel().MetaData(),
                          graph.GetModel().ModelPath(),  // use source model path so that external initializers can find the data file path
@@ -882,9 +888,7 @@ static Status CreateEpContextModel(const ExecutionProviders& execution_providers
 
   ModelSavingOptions model_saving_options{ini_size_threshold};
 
-  if (ep_context_gen_options.output_model_buffer_ptr != nullptr &&
-      ep_context_gen_options.output_model_buffer_size_ptr != nullptr &&
-      ep_context_gen_options.output_model_buffer_allocator != nullptr) {
+  if (saving_to_buffer) {
     ORT_RETURN_IF_ERROR(ep_context_model.MainGraph().Resolve());
     // TODO(adrianlizarraga): Investigate if we can make this more memory efficient.
     // May be able to use allocator to directly allocate the ModelProto to avoid a copy.

onnxruntime/core/providers/qnn/builder/onnx_ctx_model_helper.cc

@@ -247,6 +247,13 @@ Status CreateEPContextNodes(Model* model,
         } else {
           context_bin_path = context_model_path;
         }
+
+        if (context_bin_path.empty()) {
+          // Context bin path is empty, so just use the graph name (e.g., "QNNExecutionProvider_QNN_13728744673520368385_2_0").
+          // This happens if both the input model and output model are stored in buffers (i.e., there are no paths).
+          context_bin_path = ToPathString(graph_name);
+        }
+
         context_bin_path = context_bin_path + ToPathString("_qnn.bin");
         context_cache_name = std::filesystem::path(context_bin_path).filename().string();
 

onnxruntime/core/session/utils.cc

@@ -52,13 +52,18 @@ OrtStatus* CreateSessionAndLoadModel(_In_ const OrtSessionOptions* options,
 
   // If ep.context_enable is set, then ep.context_file_path is expected, otherwise ORT don't know where to generate the _ctx.onnx file
   if (options && model_path == nullptr) {
-    auto ep_context_enable = options->value.config_options.GetConfigEntry(kOrtSessionOptionEpContextEnable);
-    auto ep_context_file_path = options->value.config_options.GetConfigEntry(kOrtSessionOptionEpContextFilePath);
-    if (ep_context_enable.has_value() && ep_context_enable.value() == "1" && (!ep_context_file_path.has_value() || (ep_context_file_path.has_value() && ep_context_file_path.value().empty()))) {
+    EpContextModelGenerationOptions ep_ctx_gen_options = options->value.GetEpContextGenerationOptions();
+
+    // This is checked by the OrtCompileApi's CompileModel() function, but we check again here in case
+    // the user used the older SessionOptions' configuration entries to generate a compiled model.
+    if (ep_ctx_gen_options.enable &&
+        ep_ctx_gen_options.output_model_file_path.empty() &&
+        ep_ctx_gen_options.output_model_buffer_ptr == nullptr) {
       return OrtApis::CreateStatus(ORT_FAIL,
-                                   "CreateSessionFromArray is called with ep.context_enable enabled but an \
-empty ep.context_file_path. The system does not know where to generate the \
-EP context model. Please specify a valid ep.context_file_path.");
+                                   "Inference session was configured with EPContext model generation enabled but "
+                                   "without a valid location (e.g., file or buffer) for the output model. "
+                                   "Please specify a valid ep.context_file_path via SessionOption configs "
+                                   "or use the OrtCompileApi to compile a model to a file or buffer.");
     }
   }
 

onnxruntime/test/providers/qnn/qnn_ep_context_test.cc

@@ -436,6 +436,75 @@ TEST_F(QnnHTPBackendTests, CompileApi_FromSessionOptions_OutputModelBuffer) {
 
   // Check that the compiled model has the expected number of EPContext nodes.
   CheckEpContextNodeCounts(output_model_buffer, output_model_buffer_size, 2, 2);
+  allocator.Free(output_model_buffer);
+}
+
+// Test using the CompileModel() API with settings:
+//   - input model from buffer
+//   - save output model to buffer
+//   - test enabling AND disabling embed mode for context binary in EPContext node attributes
+TEST_F(QnnHTPBackendTests, CompileApi_FromSessionOptions_InputAndOutputModelsInBuffers) {
+  // Create a test model and serialize it to a buffer.
+  TestModel test_model;
+  CreateTestModel(BuildGraphWithQAndNonQ(false), 21, logging::Severity::kERROR, test_model);
+  std::string model_data = test_model.Serialize();
+
+  // Initialize session options with QNN EP
+  Ort::SessionOptions session_options;
+  ProviderOptions provider_options;
+  provider_options["backend_type"] = "htp";
+  provider_options["offload_graph_io_quantization"] = "0";
+  session_options.AppendExecutionProvider("QNN", provider_options);
+
+  Ort::AllocatorWithDefaultOptions allocator;
+
+  // Test embed mode enabled.
+  {
+    void* output_model_buffer = nullptr;
+    size_t output_model_buffer_size = 0;
+
+    // Create model compilation options from the session options.
+    Ort::ModelCompilationOptions compile_options(*ort_env, session_options);
+    compile_options.SetInputModelFromBuffer(reinterpret_cast<const void*>(model_data.data()), model_data.size());
+    compile_options.SetOutputModelBuffer(allocator, &output_model_buffer, &output_model_buffer_size);
+    compile_options.SetEpContextEmbedMode(true);
+
+    // Compile the model.
+    Ort::Status status = Ort::CompileModel(*ort_env, compile_options);
+    ASSERT_TRUE(status.IsOK()) << status.GetErrorMessage();
+
+    // Make sure the compiled model was saved to the buffer.
+    ASSERT_TRUE(output_model_buffer != nullptr);
+    ASSERT_TRUE(output_model_buffer_size > 0);
+
+    // Check that the compiled model has the expected number of EPContext nodes.
+    CheckEpContextNodeCounts(output_model_buffer, output_model_buffer_size, 2, 2);
+    allocator.Free(output_model_buffer);
+  }
+
+  // Test embed mode disabled.
+  {
+    void* output_model_buffer = nullptr;
+    size_t output_model_buffer_size = 0;
+
+    // Create model compilation options from the session options.
+    Ort::ModelCompilationOptions compile_options(*ort_env, session_options);
+    compile_options.SetInputModelFromBuffer(reinterpret_cast<const void*>(model_data.data()), model_data.size());
+    compile_options.SetOutputModelBuffer(allocator, &output_model_buffer, &output_model_buffer_size);
+    compile_options.SetEpContextEmbedMode(false);
+
+    // Compile the model.
+    Ort::Status status = Ort::CompileModel(*ort_env, compile_options);
+    ASSERT_TRUE(status.IsOK()) << status.GetErrorMessage();
+
+    // Make sure the compiled model was saved to the buffer.
+    ASSERT_TRUE(output_model_buffer != nullptr);
+    ASSERT_TRUE(output_model_buffer_size > 0);
+
+    // Check that the compiled model has the expected number of EPContext nodes.
+    CheckEpContextNodeCounts(output_model_buffer, output_model_buffer_size, 2, 2);
+    allocator.Free(output_model_buffer);
+  }
 }
 
 // Test using the CompileModel() API with settings:
@@ -485,6 +554,7 @@ TEST_F(QnnHTPBackendTests, CompileApi_FromSessionOptions_OutputModelBuffer_Outpu
 
   // Check that the compiled model has the expected number of EPContext nodes.
   CheckEpContextNodeCounts(output_model_buffer, output_model_buffer_size, 2, 2);
+  allocator.Free(output_model_buffer);
 }
 
 // Test that models with 1 non-quantized FusedMatMul node and 1 quantized Add node can still generate the context binary
@@ -1566,7 +1636,7 @@ TEST_F(QnnHTPBackendTests, LoadFromArrayWithQnnEpContextGenPathValidation) {
   ORT_CATCH(const std::exception& e) {
     ORT_HANDLE_EXCEPTION([&e]() {
       std::string e_message1(std::string(e.what()));
-      ASSERT_TRUE(e_message1.find("Please specify a valid ep.context_file_path.") != std::string::npos);
+      ASSERT_TRUE(e_message1.find("Please specify a valid ep.context_file_path") != std::string::npos);
     });
   }
 
@@ -1577,7 +1647,7 @@ TEST_F(QnnHTPBackendTests, LoadFromArrayWithQnnEpContextGenPathValidation) {
   ORT_CATCH(const std::exception& ex) {
     ORT_HANDLE_EXCEPTION([&ex]() {
       std::string e_message2(std::string(ex.what()));
-      ASSERT_TRUE(e_message2.find("Please specify a valid ep.context_file_path.") != std::string::npos);
+      ASSERT_TRUE(e_message2.find("Please specify a valid ep.context_file_path") != std::string::npos);
     });
   }
 }