From 82036b04978b7930185996a70d2146c2895469ea Mon Sep 17 00:00:00 2001
From: Changming Sun <chasun@microsoft.com>
Date: Mon, 29 Jul 2024 21:59:16 -0700
Subject: [PATCH] Remove references to the outdated CUDA EP factory method
 (#21549)

The function "OrtSessionOptionsAppendExecutionProvider_CUDA" is
deprecated.
---
 .../global_thread_pools/test_inference.cc     |  4 +++-
 onnxruntime/test/shared_lib/test_inference.cc | 20 ++++++++++++++-----
 .../test/shared_lib/test_model_loading.cc     |  5 +++--
 3 files changed, 21 insertions(+), 8 deletions(-)

diff --git a/onnxruntime/test/global_thread_pools/test_inference.cc b/onnxruntime/test/global_thread_pools/test_inference.cc
index f553682975f11..c6d958536f488 100644
--- a/onnxruntime/test/global_thread_pools/test_inference.cc
+++ b/onnxruntime/test/global_thread_pools/test_inference.cc
@@ -74,7 +74,9 @@ static Ort::Session GetSessionObj(Ort::Env& env, T model_uri, int provider_type)
 
   if (provider_type == 1) {
 #ifdef USE_CUDA
-    Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0));
+    OrtCUDAProviderOptionsV2* options;
+    Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+    session_options.AppendExecutionProvider_CUDA_V2(*options);
     std::cout << "Running simple inference with cuda provider" << std::endl;
 #else
     return Ort::Session(nullptr);
diff --git a/onnxruntime/test/shared_lib/test_inference.cc b/onnxruntime/test/shared_lib/test_inference.cc
index 52491a179c2ce..7a33bf8a527cd 100644
--- a/onnxruntime/test/shared_lib/test_inference.cc
+++ b/onnxruntime/test/shared_lib/test_inference.cc
@@ -1959,7 +1959,9 @@ TEST(CApiTest, get_allocator_cpu) {
 #ifdef USE_CUDA
 TEST(CApiTest, get_allocator_cuda) {
   Ort::SessionOptions session_options;
-  Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0));
+  OrtCUDAProviderOptionsV2* options;
+  Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+  session_options.AppendExecutionProvider_CUDA_V2(*options);
   Ort::Session session(*ort_env, NAMED_AND_ANON_DIM_PARAM_URI, session_options);
 
   Ort::MemoryInfo info_cuda("Cuda", OrtAllocatorType::OrtArenaAllocator, 0, OrtMemTypeDefault);
@@ -2076,7 +2078,9 @@ TEST(CApiTest, io_binding_cuda) {
 #ifdef USE_TENSORRT
   Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_Tensorrt(session_options, 0));
 #else
-  Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0));
+  OrtCUDAProviderOptionsV2* options;
+  Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+  session_options.AppendExecutionProvider_CUDA_V2(*options);
 #endif
   Ort::Session session(*ort_env, MODEL_URI, session_options);
 
@@ -3438,7 +3442,9 @@ TEST(CApiTest, AllocateInitializersFromNonArenaMemory) {
   Ort::SessionOptions session_options;
 
 #ifdef USE_CUDA
-  Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0));
+  OrtCUDAProviderOptionsV2* options;
+  Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+  session_options.AppendExecutionProvider_CUDA_V2(*options);
 #else
   // arena is enabled but the sole initializer will still be allocated from non-arena memory
   Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CPU(session_options, 1));
@@ -3890,7 +3896,9 @@ TEST(CApiTest, GitHubIssue10179) {
     try {
       const auto* model_path = MODEL_URI;
       Ort::SessionOptions session_options{};
-      Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0));
+      OrtCUDAProviderOptionsV2* options;
+      Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+      session_options.AppendExecutionProvider_CUDA_V2(*options);
       Ort::Session session{*ort_env, model_path, session_options};
     } catch (const std::exception& e) {
       std::cerr << "exception: " << e.what() << "\n";
@@ -3920,7 +3928,9 @@ TEST(CApiTest, GitHubIssue10179) {
 TEST(CApiTest, TestCudaMemcpyToHostWithSequenceTensors) {
   const auto* model_path = SEQUENCE_MODEL_URI_2;
   Ort::SessionOptions session_options{};
-  Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(session_options, 0));
+  OrtCUDAProviderOptionsV2* options;
+  Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+  session_options.AppendExecutionProvider_CUDA_V2(*options);
   Ort::Session session{*ort_env, model_path, session_options};
 
   Ort::MemoryInfo info("Cpu", OrtDeviceAllocator, 0, OrtMemTypeDefault);
diff --git a/onnxruntime/test/shared_lib/test_model_loading.cc b/onnxruntime/test/shared_lib/test_model_loading.cc
index b7f6f7f4b9a77..5694398b9cb10 100644
--- a/onnxruntime/test/shared_lib/test_model_loading.cc
+++ b/onnxruntime/test/shared_lib/test_model_loading.cc
@@ -60,8 +60,9 @@ TEST(CApiTest, model_from_array) {
   create_session(so);
 
 #ifdef USE_CUDA
-  // test with CUDA provider when using onnxruntime as dll
-  Ort::ThrowOnError(OrtSessionOptionsAppendExecutionProvider_CUDA(so, 0));
+  OrtCUDAProviderOptionsV2* options;
+  Ort::ThrowOnError(Ort::GetApi().CreateCUDAProviderOptions(&options));
+  so.AppendExecutionProvider_CUDA_V2(*options);
   create_session(so);
 #endif
 }