Missing cuda_provider_factory.h file in release for windows gpu

[dgarlor]

Describe the bug
Using the release onnxruntime-win-x64-gpu-0.4.0.zip.
Following the example to execute a model using the C API, I couldn't activate the CUDA provider because cuda_provider_factory.h file is missing.

I found a workaround compiling the project myself and copying the generated file into my include folder (also modifying the #include inside the file)

Urgency
No urgency as I found the workaround but it would be easier for future releases.

System information

• Windows 10
• ONNX Runtime installed from (source or binary): binary
• ONNX Runtime version: 0.4.0
• Python version: 3.7 (irrelevant)
• Visual Studio version (if applicable): 2017
• CUDA/cuDNN version: 9.1 / 7.1

Thanks and good work

[hariharans29]

Thanks for reporting this.

CC: @pranavsharma @snnn

[shahasad]

resolved by #1461

[OpDaSo]

I have the same problem with the release version onnxruntime-win-x64-gpu-1.11.0.zip

[5p4k]

Also absent from onnxruntime-linux-x64-gpu-1.11.0.

[snnn]

It is not relevant any more .

#include "onnxruntime_c_api.h"

        OrtCUDAProviderOptionsV2* cuda_options = nullptr;
        api->CreateCUDAProviderOptions(&cuda_options));
        std::unique_ptr<OrtCUDAProviderOptionsV2, decltype(&OrtApis::ReleaseCUDAProviderOptions)> rel_cuda_options(
            cuda_options, &OrtApis::ReleaseCUDAProviderOptions);
        api->SessionOptionsAppendExecutionProvider_CUDA_V2(ortso, cuda_options));

[OpDaSo]

What is the type of the variable "api"?

I want to run the mnist example (https://github.com/microsoft/onnxruntime-inference-examples/tree/main/c_cxx/MNIST).
Unfortunately, I get the error "Cannot open source file: cuda_provider_factory.h" in the file providers.h line 8.

Do you have an example for using onnxruntime with CUDA in C++?

[snnn]

const OrtApi* api = OrtGetApiBase()->GetApi(ORT_API_VERSION);

[jywu-msft]

the examples need to be updated to not include cuda_provider_factory.h