From 7bf00902060f8c08aed28dc82bef461d3560f656 Mon Sep 17 00:00:00 2001
From: Ryan Hill <ryanhill@microsoft.com>
Date: Wed, 21 Apr 2021 01:50:59 -0700
Subject: [PATCH] Test disabling cast

---
 onnxruntime/core/providers/cuda/cuda_execution_provider.cc | 2 ++
 onnxruntime/test/util/default_providers.cc                 | 1 +
 2 files changed, 3 insertions(+)

diff --git a/onnxruntime/core/providers/cuda/cuda_execution_provider.cc b/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
index 41096768b4..4a5ad51fe0 100644
--- a/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
+++ b/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
@@ -1329,6 +1329,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, float, ReduceLogSumExp)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, double, ReduceLogSumExp)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 1, 10, MLFloat16, ReduceLogSumExp)>,
+#if 0
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, 8, float, Cast)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, 8, double, Cast)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 6, 8, MLFloat16, Cast)>,
@@ -1353,6 +1354,7 @@ static Status RegisterCudaKernels(KernelRegistry& kernel_registry) {
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, 12, uint32_t, Cast)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, 12, uint64_t, Cast)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 9, 12, bool, Cast)>,
+#endif
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 2, 10, float, Pad)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 2, 10, double, Pad)>,
     BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 2, 10, MLFloat16, Pad)>,
diff --git a/onnxruntime/test/util/default_providers.cc b/onnxruntime/test/util/default_providers.cc
index a28ab03dcc..1d8276630c 100644
--- a/onnxruntime/test/util/default_providers.cc
+++ b/onnxruntime/test/util/default_providers.cc
@@ -71,6 +71,7 @@ std::unique_ptr<IExecutionProvider> DefaultOpenVINOExecutionProvider() {
 std::unique_ptr<IExecutionProvider> DefaultCudaExecutionProvider() {
 #ifdef USE_CUDA
   OrtCUDAProviderOptions provider_options{};
+  provider_options.do_copy_in_default_stream = true;
   if (auto factory = CreateExecutionProviderFactory_Cuda(&provider_options))
     return factory->CreateProvider();
 #endif