How to install a custom build of onnxruntime on Windows?

[ae137]

I am trying to create a custom build of onnxruntime for C++ with CUDA execution provider and install it on Windows (the encountered issue is independent of the custom build). I have built the library as described here, using the tag v1.10.0, CUDA 11.4, CMake 3.21.1 and MSVC 19.29. Building and testing works fine.

However, when I switch to build/Windows/Release and type cmake --install ., the installation folder seems to be incomplete: There is a folder include/onnxruntime/core/providers/cpu, but none for cuda. In the source directory, there is a corresponding folder onnxruntime/core/providers, and I wonder which of its content should end up in the installation directory?

I have found a somewhat similar discussion for Linux here, but there are no direct equivalents for the scripts on Windows. Is it possible to bundle the installation files with CMake, or required to use Nuget?

[ytaous]

#11807